text
stringlengths 7
318k
| id
stringlengths 14
166
| metadata
dict | __index_level_0__
int64 0
439
|
|---|---|---|---|
<jupyter_start><jupyter_text>Going Production: Auto-scale Hugging Face Transformer Endpoints with Amazon SageMaker Welcome to this getting started guide, we will use the new Hugging Face Inference DLCs and Amazon SageMaker Python SDK to deploy a transformer model for real-time inference. In this example we are going to deploy a trained Hugging Face Transformer model on to SageMaker for inference.<jupyter_code>!pip install "sagemaker>=2.66.2" --upgrade<jupyter_output><empty_output><jupyter_text>Deploy one of the 15 000+ Hugging Face Transformers to Amazon SageMaker for InferenceTo deploy a model directly from the Hub to SageMaker we need to define 2 environment variables when creating the `HuggingFaceModel` . We need to define:- `HF_MODEL_ID`: defines the model id, which will be automatically loaded from [huggingface.co/models](http://huggingface.co/models) when creating or SageMaker Endpoint. The ð€ Hub provides +10 000 models all available through this environment variable.- `HF_TASK`: defines the task for the used ð€ Transformers pipeline. A full list of tasks can be find [here](https://huggingface.co/transformers/main_classes/pipelines.html).<jupyter_code>import sagemaker
import boto3
try:
role = sagemaker.get_execution_role()
except ValueError:
iam = boto3.client('iam')
role = iam.get_role(RoleName='sagemaker_execution_role')['Role']['Arn']
print(f"sagemaker role arn: {role}")
from sagemaker.huggingface import HuggingFaceModel
from uuid import uuid4
import sagemaker
# Hub Model configuration. https://huggingface.co/models
hub = {
'HF_MODEL_ID':'yiyanghkust/finbert-tone', # model_id from hf.co/models
'HF_TASK':'text-classification' # NLP task you want to use for predictions
}
# endpoint name
endpoint_name=f'{hub["HF_MODEL_ID"].split("/")[1]}-{str(uuid4())}' # model and endpoint name
# create Hugging Face Model Class
huggingface_model = HuggingFaceModel(
env=hub,
role=role, # iam role with permissions to create an Endpoint
name=endpoint_name, # model and endpoint name
transformers_version="4.26", # transformers version used
pytorch_version="1.13", # pytorch version used
py_version="py39", # python version of the DLC
)
# deploy model to SageMaker Inference
predictor = huggingface_model.deploy(
initial_instance_count=1,
instance_type="ml.c5.large"
)
# get aws region for dashboards
aws_region = predictor.sagemaker_session.boto_region_name<jupyter_output>-----!<jupyter_text>**Architecture**The [Hugging Face Inference Toolkit for SageMaker](https://github.com/aws/sagemaker-huggingface-inference-toolkit) is an open-source library for serving Hugging Face transformer models on SageMaker. It utilizes the SageMaker Inference Toolkit for starting up the model server, which is responsible for handling inference requests. The SageMaker Inference Toolkit uses [Multi Model Server (MMS)](https://github.com/awslabs/multi-model-server) for serving ML models. It bootstraps MMS with a configuration and settings that make it compatible with SageMaker and allow you to adjust important performance parameters, such as the number of workers per model, depending on the needs of your scenario.**Deploying a model using SageMaker hosting services is a three-step process:**1. **Create a model in SageMaker** âBy creating a model, you tell SageMaker where it can find the model components. 2. **Create an endpoint configuration for an HTTPS endpoint** âYou specify the name of one or more models in production variants and the ML compute instances that you want SageMaker to launch to host each production variant.3. **Create an HTTPS endpoint** âProvide the endpoint configuration to SageMaker. The service launches the ML compute instances and deploys the model or models as specified in the configuration<jupyter_code># example request, you always need to define "inputs"
data = {
"inputs": "There is a shortage of capital for project SageMaker. We need extra financing"
}
# request
predictor.predict(data)
for i in range(500):
predictor.predict(data)<jupyter_output><empty_output><jupyter_text>Model Monitoring<jupyter_code>print(f"https://console.aws.amazon.com/cloudwatch/home?region={aws_region}#metricsV2:graph=~(metrics~(~(~'AWS*2fSageMaker~'ModelLatency~'EndpointName~'finbert-tone-73d26f97-9376-4b3f-9334-a2-2021-10-29-12-18-52-365~'VariantName~'AllTraffic))~view~'timeSeries~stacked~false~start~'-PT15M~end~'P0D~region~'{aws_region}~stat~'SampleCount~period~30);query=~'*7bAWS*2fSageMaker*2cEndpointName*2cVariantName*7d*20{predictor.endpoint_name}")<jupyter_output><empty_output><jupyter_text>Auto Scaling your Model[Amazon SageMaker](https://aws.amazon.com/sagemaker/) is a fully managed service that provides every developer and data scientist with the ability to quickly build, train, and deploy machine learning (ML) models at scale.Autoscaling is an out-of-the-box feature that monitors your workloads and dynamically adjusts the capacity to maintain steady and predictable performance at the possible lowest cost.The following diagram is a sample architecture that showcases how a model is served as a endpoint with autoscaling enabled. Reference Blog post [Configuring autoscaling inference endpoints in Amazon SageMaker](https://aws.amazon.com/de/blogs/machine-learning/configuring-autoscaling-inference-endpoints-in-amazon-sagemaker/) Configure Autoscaling for our EndpointYou can define minimum, desired, and maximum number of instances per endpoint and, based on the autoscaling configurations, instances are managed dynamically. The following diagram illustrates this architecture. AWS offers many different [ways to auto-scale your endpoints](https://docs.aws.amazon.com/autoscaling/application/userguide/application-auto-scaling-target-tracking.html). One of them Simple-Scaling, where you scale the instance capacity based on `CPUUtilization` of the instances or `SageMakerVariantInvocationsPerInstance`. In this example we are going to use `SageMakerVariantInvocationsPerInstance` to auto-scale our Endpoint<jupyter_code>import boto3
# Let us define a client to play with autoscaling options
asg_client = boto3.client('application-autoscaling') # Common class representing Application Auto Scaling for SageMaker amongst other services
# he resource type is variant and the unique identifier is the resource ID.
# Example: endpoint/my-bert-fine-tuned/variant/AllTraffic .
resource_id=f"endpoint/{predictor.endpoint_name}/variant/AllTraffic"
# scaling configuration
response = asg_client.register_scalable_target(
ServiceNamespace='sagemaker', #
ResourceId=resource_id,
ScalableDimension='sagemaker:variant:DesiredInstanceCount',
MinCapacity=1,
MaxCapacity=4
)<jupyter_output><empty_output><jupyter_text>Create Scaling Policy with configuration details, e.g. `TargetValue` when the instance should be scaled.<jupyter_code>response = asg_client.put_scaling_policy(
PolicyName=f'Request-ScalingPolicy-{predictor.endpoint_name}',
ServiceNamespace='sagemaker',
ResourceId=resource_id,
ScalableDimension='sagemaker:variant:DesiredInstanceCount',
PolicyType='TargetTrackingScaling',
TargetTrackingScalingPolicyConfiguration={
'TargetValue': 10.0, # Threshold
'PredefinedMetricSpecification': {
'PredefinedMetricType': 'SageMakerVariantInvocationsPerInstance',
},
'ScaleInCooldown': 300, # duration until scale in
'ScaleOutCooldown': 60 # duration between scale out
}
)<jupyter_output><empty_output><jupyter_text>stress test the endpoint with threaded requests<jupyter_code>import time
request_duration_in_seconds = 4*65
end_time = time.time() + request_duration_in_seconds
print(f"test will run {request_duration_in_seconds} seconds")
while time.time() < end_time:
predictor.predict(data)<jupyter_output><empty_output><jupyter_text>Monitor the `InvocationsPerInstance` in cloudwatch<jupyter_code>print(f"https://console.aws.amazon.com/cloudwatch/home?region={aws_region}#metricsV2:graph=~(metrics~(~(~'AWS*2fSageMaker~'InvocationsPerInstance~'EndpointName~'{predictor.endpoint_name}~'VariantName~'AllTraffic))~view~'timeSeries~stacked~false~region~'{aws_region}~start~'-PT15M~end~'P0D~stat~'SampleCount~period~60);query=~'*7bAWS*2fSageMaker*2cEndpointName*2cVariantName*7d*20{predictor.endpoint_name}")<jupyter_output><empty_output><jupyter_text>check the endpoint instance_count number<jupyter_code>bt_sm = boto3.client('sagemaker')
response = bt_sm.describe_endpoint(EndpointName=predictor.endpoint_name)
print(f"Endpoint {response['EndpointName']} has \nCurrent Instance Count: {response['ProductionVariants'][0]['CurrentInstanceCount']}\nWith a desired instance count of {response['ProductionVariants'][0]['DesiredInstanceCount']}")<jupyter_output>Endpoint finbert-tone-73d26f97-9376-4b3f-9334-a2-2021-10-29-12-18-52-365 has
Current Instance Count: 4
With a desired instance count of 4<jupyter_text>Clean up<jupyter_code># delete endpoint
predictor.delete_model()
predictor.delete_endpoint()<jupyter_output><empty_output> | notebooks/sagemaker/13_deploy_and_autoscaling_transformers/sagemaker-notebook.ipynb/0 | {
"file_path": "notebooks/sagemaker/13_deploy_and_autoscaling_transformers/sagemaker-notebook.ipynb",
"repo_id": "notebooks",
"token_count": 2793
} | 155 |
import os
from transformers import AutoConfig, AutoTokenizer
import torch
import torch.neuron
# To use one neuron core per worker
os.environ["NEURON_RT_NUM_CORES"] = "1"
# saved weights name
AWS_NEURON_TRACED_WEIGHTS_NAME = "neuron_model.pt"
def model_fn(model_dir):
# load tokenizer and neuron model from model_dir
tokenizer = AutoTokenizer.from_pretrained(model_dir)
model = torch.jit.load(os.path.join(model_dir, AWS_NEURON_TRACED_WEIGHTS_NAME))
model_config = AutoConfig.from_pretrained(model_dir)
return model, tokenizer, model_config
def predict_fn(data, model_tokenizer_model_config):
# destruct model, tokenizer and model config
model, tokenizer, model_config = model_tokenizer_model_config
# create embeddings for inputs
inputs = data.pop("inputs", data)
embeddings = tokenizer(
inputs,
return_tensors="pt",
max_length=model_config.traced_sequence_length,
padding="max_length",
truncation=True,
)
# convert to tuple for neuron model
neuron_inputs = tuple(embeddings.values())
# run prediciton
with torch.no_grad():
predictions = model(*neuron_inputs)[0]
scores = torch.nn.Softmax(dim=1)(predictions)
# return dictonary, which will be json serializable
return [{"label": model_config.id2label[item.argmax().item()], "score": item.max().item()} for item in scores]
| notebooks/sagemaker/18_inferentia_inference/code/inference.py/0 | {
"file_path": "notebooks/sagemaker/18_inferentia_inference/code/inference.py",
"repo_id": "notebooks",
"token_count": 519
} | 156 |
import os
import argparse
from transformers import (
AutoModelForCausalLM,
AutoTokenizer,
set_seed,
default_data_collator,
)
from datasets import load_from_disk
import torch
from transformers import Trainer, TrainingArguments
import torch.distributed as dist
def safe_save_model_for_hf_trainer(trainer: Trainer, tokenizer: AutoTokenizer, output_dir: str):
"""Helper method to save model for HF Trainer."""
# see: https://github.com/tatsu-lab/stanford_alpaca/issues/65
from torch.distributed.fsdp import (
FullyShardedDataParallel as FSDP,
FullStateDictConfig,
StateDictType,
)
model = trainer.model
save_policy = FullStateDictConfig(offload_to_cpu=True, rank0_only=True)
with FSDP.state_dict_type(model, StateDictType.FULL_STATE_DICT, save_policy):
cpu_state_dict = model.state_dict()
if trainer.args.should_save:
trainer._save(output_dir, state_dict=cpu_state_dict) # noqa
tokenizer.save_pretrained(output_dir)
def parse_arge():
"""Parse the arguments."""
parser = argparse.ArgumentParser()
# add model id and dataset path argument
parser.add_argument(
"--model_id",
type=str,
default="google/flan-t5-xl",
help="Model id to use for training.",
)
parser.add_argument("--dataset_path", type=str, default="lm_dataset", help="Path to dataset.")
# add training hyperparameters for epochs, batch size, learning rate, and seed
parser.add_argument("--epochs", type=int, default=3, help="Number of epochs to train for.")
parser.add_argument("--max_steps", type=int, default=None, help="Number of epochs to train for.")
parser.add_argument(
"--per_device_train_batch_size",
type=int,
default=1,
help="Batch size to use for training.",
)
parser.add_argument("--lr", type=float, default=3e-5, help="Learning rate to use for training.")
parser.add_argument("--optimizer", type=str, default="adamw_hf", help="Learning rate to use for training.")
parser.add_argument("--seed", type=int, default=42, help="Seed to use for training.")
parser.add_argument(
"--gradient_checkpointing",
type=bool,
default=True,
help="Path to deepspeed config file.",
)
parser.add_argument(
"--bf16",
type=bool,
default=True if torch.cuda.get_device_capability()[0] == 8 else False,
help="Whether to use bf16.",
)
parser.add_argument("--fsdp", type=str, default=None, help="Whether to use fsdp.")
parser.add_argument(
"--fsdp_transformer_layer_cls_to_wrap",
type=str,
default=None,
help="Which transformer layer to wrap with fsdp.",
)
args = parser.parse_known_args()
return args
def training_function(args):
# set seed
set_seed(args.seed)
dataset = load_from_disk(args.dataset_path)
# load model from the hub
model = AutoModelForCausalLM.from_pretrained(
args.model_id,
use_cache=False if args.gradient_checkpointing else True, # this is needed for gradient checkpointing
)
tokenizer = AutoTokenizer.from_pretrained(args.model_id)
# Define training args
output_dir = "/tmp"
training_args = TrainingArguments(
output_dir=output_dir,
overwrite_output_dir=True,
per_device_train_batch_size=args.per_device_train_batch_size,
bf16=args.bf16, # Use BF16 if available
learning_rate=args.lr,
num_train_epochs=args.epochs,
gradient_checkpointing=args.gradient_checkpointing,
# logging strategies
logging_dir=f"{output_dir}/logs",
logging_strategy="steps",
logging_steps=10,
save_strategy="no",
optim=args.optimizer,
ddp_timeout=7200,
fsdp=args.fsdp,
fsdp_transformer_layer_cls_to_wrap=args.fsdp_transformer_layer_cls_to_wrap,
)
# Create Trainer instance
trainer = Trainer(
model=model,
args=training_args,
train_dataset=dataset,
data_collator=default_data_collator,
)
# Start training
trainer.train()
print("Training done!")
# save model and tokenizer for easy inference
safe_save_model_for_hf_trainer(trainer, tokenizer, "/opt/ml/model/")
dist.barrier()
def main():
args, _ = parse_arge()
training_function(args)
if __name__ == "__main__":
main()
| notebooks/sagemaker/25_pytorch_fsdp_model_parallelism/scripts/run_clm.py/0 | {
"file_path": "notebooks/sagemaker/25_pytorch_fsdp_model_parallelism/scripts/run_clm.py",
"repo_id": "notebooks",
"token_count": 1807
} | 157 |
- title: Get started
sections:
- local: index
title: ð€ PEFT
- local: quicktour
title: Quicktour
- local: install
title: Installation
- title: Tutorial
sections:
- local: tutorial/peft_model_config
title: Configurations and models
- local: tutorial/peft_integrations
title: Integrations
- title: PEFT method guides
sections:
- local: task_guides/prompt_based_methods
title: Prompt-based methods
- title: LoRA
sections:
- local: task_guides/image_classification_lora
title: Image classification
- local: task_guides/semantic_segmentation_lora
title: Semantic segmentation
- local: task_guides/token-classification-lora
title: Token classification
- local: task_guides/semantic-similarity-lora
title: Semantic similarity
- local: task_guides/int8-asr
title: int8 training for automatic speech recognition
- local: task_guides/dreambooth_lora
title: DreamBooth
- title: Developer guides
sections:
- local: developer_guides/quantization
title: Quantization
- local: developer_guides/lora
title: LoRA
- local: developer_guides/custom_models
title: Working with custom models
- local: developer_guides/low_level_api
title: PEFT low level API
- local: developer_guides/mixed_models
title: Mixing different adapter types
- local: developer_guides/contributing
title: Contributing to PEFT
- local: developer_guides/troubleshooting
title: Troubleshooting
- title: ð€ Accelerate integrations
sections:
- local: accelerate/deepspeed-zero3-offload
title: DeepSpeed
- local: accelerate/fsdp
title: Fully Sharded Data Parallel
- title: Conceptual guides
sections:
- local: conceptual_guides/adapter
title: Adapters
- local: conceptual_guides/prompting
title: Soft prompts
- local: conceptual_guides/ia3
title: IA3
- sections:
- sections:
- local: package_reference/auto_class
title: AutoPeftModel
- local: package_reference/peft_model
title: PEFT model
- local: package_reference/peft_types
title: PEFT types
- local: package_reference/config
title: Configuration
- local: package_reference/tuners
title: Tuner
title: Main classes
- sections:
- local: package_reference/adalora
title: AdaLoRA
- local: package_reference/ia3
title: IA3
- local: package_reference/llama_adapter
title: Llama-Adapter
- local: package_reference/loha
title: LoHa
- local: package_reference/lokr
title: LoKr
- local: package_reference/lora
title: LoRA
- local: package_reference/adapter_utils
title: LyCORIS
- local: package_reference/multitask_prompt_tuning
title: Multitask Prompt Tuning
- local: package_reference/oft
title: OFT
- local: package_reference/poly
title: Polytropon
- local: package_reference/p_tuning
title: P-tuning
- local: package_reference/prefix_tuning
title: Prefix tuning
- local: package_reference/prompt_tuning
title: Prompt tuning
title: Adapters
title: API reference
| peft/docs/source/_toctree.yml/0 | {
"file_path": "peft/docs/source/_toctree.yml",
"repo_id": "peft",
"token_count": 1139
} | 158 |
<jupyter_start><jupyter_code>from transformers import AutoModelForCausalLM
from peft import get_peft_config, get_peft_model, PromptTuningInit, PromptTuningConfig, TaskType, PeftType
import torch
from datasets import load_dataset
import os
from transformers import AutoTokenizer
from torch.utils.data import DataLoader
from transformers import default_data_collator, get_linear_schedule_with_warmup
from tqdm import tqdm
from datasets import load_dataset
device = "cuda"
model_name_or_path = "bigscience/bloomz-560m"
tokenizer_name_or_path = "bigscience/bloomz-560m"
peft_config = PromptTuningConfig(
task_type=TaskType.CAUSAL_LM,
prompt_tuning_init=PromptTuningInit.TEXT,
num_virtual_tokens=8,
prompt_tuning_init_text="Classify if the tweet is a complaint or not:",
tokenizer_name_or_path=model_name_or_path,
)
dataset_name = "twitter_complaints"
checkpoint_name = f"{dataset_name}_{model_name_or_path}_{peft_config.peft_type}_{peft_config.task_type}_v1.pt".replace(
"/", "_"
)
text_column = "Tweet text"
label_column = "text_label"
max_length = 64
lr = 3e-2
num_epochs = 50
batch_size = 8
from datasets import load_dataset
dataset = load_dataset("ought/raft", dataset_name)
classes = [k.replace("_", " ") for k in dataset["train"].features["Label"].names]
print(classes)
dataset = dataset.map(
lambda x: {"text_label": [classes[label] for label in x["Label"]]},
batched=True,
num_proc=1,
)
print(dataset)
dataset["train"][0]
# data preprocessing
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
if tokenizer.pad_token_id is None:
tokenizer.pad_token_id = tokenizer.eos_token_id
target_max_length = max([len(tokenizer(class_label)["input_ids"]) for class_label in classes])
print(target_max_length)
def preprocess_function(examples):
batch_size = len(examples[text_column])
inputs = [f"{text_column} : {x} Label : " for x in examples[text_column]]
targets = [str(x) for x in examples[label_column]]
model_inputs = tokenizer(inputs)
labels = tokenizer(targets, add_special_tokens=False) # don't add bos token because we concatenate with inputs
for i in range(batch_size):
sample_input_ids = model_inputs["input_ids"][i]
label_input_ids = labels["input_ids"][i] + [tokenizer.eos_token_id]
# print(i, sample_input_ids, label_input_ids)
model_inputs["input_ids"][i] = sample_input_ids + label_input_ids
labels["input_ids"][i] = [-100] * len(sample_input_ids) + label_input_ids
model_inputs["attention_mask"][i] = [1] * len(model_inputs["input_ids"][i])
# print(model_inputs)
for i in range(batch_size):
sample_input_ids = model_inputs["input_ids"][i]
label_input_ids = labels["input_ids"][i]
model_inputs["input_ids"][i] = [tokenizer.pad_token_id] * (
max_length - len(sample_input_ids)
) + sample_input_ids
model_inputs["attention_mask"][i] = [0] * (max_length - len(sample_input_ids)) + model_inputs[
"attention_mask"
][i]
labels["input_ids"][i] = [-100] * (max_length - len(sample_input_ids)) + label_input_ids
model_inputs["input_ids"][i] = torch.tensor(model_inputs["input_ids"][i][:max_length])
model_inputs["attention_mask"][i] = torch.tensor(model_inputs["attention_mask"][i][:max_length])
labels["input_ids"][i] = torch.tensor(labels["input_ids"][i][:max_length])
model_inputs["labels"] = labels["input_ids"]
return model_inputs
processed_datasets = dataset.map(
preprocess_function,
batched=True,
num_proc=1,
remove_columns=dataset["train"].column_names,
load_from_cache_file=False,
desc="Running tokenizer on dataset",
)
train_dataset = processed_datasets["train"]
eval_dataset = processed_datasets["train"]
train_dataloader = DataLoader(
train_dataset, shuffle=True, collate_fn=default_data_collator, batch_size=batch_size, pin_memory=True
)
eval_dataloader = DataLoader(eval_dataset, collate_fn=default_data_collator, batch_size=batch_size, pin_memory=True)
def test_preprocess_function(examples):
batch_size = len(examples[text_column])
inputs = [f"{text_column} : {x} Label : " for x in examples[text_column]]
model_inputs = tokenizer(inputs)
# print(model_inputs)
for i in range(batch_size):
sample_input_ids = model_inputs["input_ids"][i]
model_inputs["input_ids"][i] = [tokenizer.pad_token_id] * (
max_length - len(sample_input_ids)
) + sample_input_ids
model_inputs["attention_mask"][i] = [0] * (max_length - len(sample_input_ids)) + model_inputs[
"attention_mask"
][i]
model_inputs["input_ids"][i] = torch.tensor(model_inputs["input_ids"][i][:max_length])
model_inputs["attention_mask"][i] = torch.tensor(model_inputs["attention_mask"][i][:max_length])
return model_inputs
test_dataset = dataset["test"].map(
test_preprocess_function,
batched=True,
num_proc=1,
remove_columns=dataset["train"].column_names,
load_from_cache_file=False,
desc="Running tokenizer on dataset",
)
test_dataloader = DataLoader(test_dataset, collate_fn=default_data_collator, batch_size=batch_size, pin_memory=True)
next(iter(test_dataloader))
next(iter(train_dataloader))
len(test_dataloader)
next(iter(test_dataloader))
# creating model
model = AutoModelForCausalLM.from_pretrained(model_name_or_path)
model = get_peft_model(model, peft_config)
model.print_trainable_parameters()
# model
# optimizer and lr scheduler
optimizer = torch.optim.AdamW(model.parameters(), lr=lr)
lr_scheduler = get_linear_schedule_with_warmup(
optimizer=optimizer,
num_warmup_steps=0,
num_training_steps=(len(train_dataloader) * num_epochs),
)
# training and evaluation
model = model.to(device)
for epoch in range(num_epochs):
model.train()
total_loss = 0
for step, batch in enumerate(tqdm(train_dataloader)):
batch = {k: v.to(device) for k, v in batch.items()}
# print(batch)
# print(batch["input_ids"].shape)
outputs = model(**batch)
loss = outputs.loss
total_loss += loss.detach().float()
loss.backward()
optimizer.step()
lr_scheduler.step()
optimizer.zero_grad()
model.eval()
eval_loss = 0
eval_preds = []
for step, batch in enumerate(tqdm(eval_dataloader)):
batch = {k: v.to(device) for k, v in batch.items()}
with torch.no_grad():
outputs = model(**batch)
loss = outputs.loss
eval_loss += loss.detach().float()
eval_preds.extend(
tokenizer.batch_decode(torch.argmax(outputs.logits, -1).detach().cpu().numpy(), skip_special_tokens=True)
)
eval_epoch_loss = eval_loss / len(eval_dataloader)
eval_ppl = torch.exp(eval_epoch_loss)
train_epoch_loss = total_loss / len(train_dataloader)
train_ppl = torch.exp(train_epoch_loss)
print(f"{epoch=}: {train_ppl=} {train_epoch_loss=} {eval_ppl=} {eval_epoch_loss=}")
model.eval()
i = 33
inputs = tokenizer(f'{text_column} : {dataset["test"][i]["Tweet text"]} Label : ', return_tensors="pt")
print(dataset["test"][i]["Tweet text"])
print(inputs)
with torch.no_grad():
inputs = {k: v.to(device) for k, v in inputs.items()}
outputs = model.generate(
input_ids=inputs["input_ids"], attention_mask=inputs["attention_mask"], max_new_tokens=10, eos_token_id=3
)
print(outputs)
print(tokenizer.batch_decode(outputs.detach().cpu().numpy(), skip_special_tokens=True))<jupyter_output>@TommyHilfiger Dramatic shopping exp. ordered 6 jeans same size (30/32) 2 fits / 2 too large / 2 too slim : same brand > different sizing
{'input_ids': tensor([[227985, 5484, 915, 2566, 226154, 126015, 5385, 259, 239364,
3396, 70823, 5853, 17, 57247, 1231, 191040, 5025, 7869,
375, 2324, 149349, 12, 415, 122321, 897, 415, 10136,
10021, 897, 415, 10136, 6497, 381, 915, 5025, 51950,
66869, 5955, 272, 20311, 77658, 915, 210]]), 'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]])}
tensor([[227985, 5484, 915, 2566, 226154, 126015, 5385, 259, 239364,
3396, 70823, 5853, 17, 57247, 1231, 191040, 5025, 7869,
375, 2324, 149349, 12, 415, 122321, 897, 415, 10136,
10021, 897, 415, 10136, [...]<jupyter_text>You can push model to hub or save model locally. - Option1: Pushing the model to Hugging Face Hub```pythonmodel.push_to_hub( f"{dataset_name}_{model_name_or_path}_{peft_config.peft_type}_{peft_config.task_type}".replace("/", "_"), token = "hf_...")```token (`bool` or `str`, *optional*): `token` is to be used for HTTP Bearer authorization when accessing remote files. If `True`, will use the token generated when running `huggingface-cli login` (stored in `~/.huggingface`). Will default to `True` if `repo_url` is not specified. Or you can get your token from https://huggingface.co/settings/token```- Or save model locally```pythonpeft_model_id = f"{dataset_name}_{model_name_or_path}_{peft_config.peft_type}_{peft_config.task_type}".replace("/", "_")model.save_pretrained(peft_model_id)```<jupyter_code># saving model
peft_model_id = f"{dataset_name}_{model_name_or_path}_{peft_config.peft_type}_{peft_config.task_type}".replace(
"/", "_"
)
model.save_pretrained(peft_model_id)
ckpt = f"{peft_model_id}/adapter_model.bin"
!du -h $ckpt
from peft import PeftModel, PeftConfig
peft_model_id = f"{dataset_name}_{model_name_or_path}_{peft_config.peft_type}_{peft_config.task_type}".replace(
"/", "_"
)
config = PeftConfig.from_pretrained(peft_model_id)
model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path)
model = PeftModel.from_pretrained(model, peft_model_id)
model.to(device)
model.eval()
i = 4
inputs = tokenizer(f'{text_column} : {dataset["test"][i]["Tweet text"]} Label : ', return_tensors="pt")
print(dataset["test"][i]["Tweet text"])
print(inputs)
with torch.no_grad():
inputs = {k: v.to(device) for k, v in inputs.items()}
outputs = model.generate(
input_ids=inputs["input_ids"], attention_mask=inputs["attention_mask"], max_new_tokens=10, eos_token_id=3
)
print(outputs)
print(tokenizer.batch_decode(outputs.detach().cpu().numpy(), skip_special_tokens=True))<jupyter_output>@greateranglia Ok thanks...
{'input_ids': tensor([[227985, 5484, 915, 2566, 14173, 2960, 29906, 387, 20706,
49337, 1369, 77658, 915, 210]]), 'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]])}
tensor([[227985, 5484, 915, 2566, 14173, 2960, 29906, 387, 20706,
49337, 1369, 77658, 915, 210, 1936, 106863, 3]],
device='cuda:0')
['Tweet text : @greateranglia Ok thanks... Label : no complaint'] | peft/examples/causal_language_modeling/peft_prompt_tuning_clm.ipynb/0 | {
"file_path": "peft/examples/causal_language_modeling/peft_prompt_tuning_clm.ipynb",
"repo_id": "peft",
"token_count": 4787
} | 159 |
import argparse
import os
from typing import Dict
import torch
from diffusers import UNet2DConditionModel
from safetensors.torch import save_file
from transformers import CLIPTextModel
from peft import PeftModel, get_peft_model_state_dict
# Default kohya_ss LoRA replacement modules
# https://github.com/kohya-ss/sd-scripts/blob/c924c47f374ac1b6e33e71f82948eb1853e2243f/networks/lora.py#L664
LORA_PREFIX_UNET = "lora_unet"
LORA_PREFIX_TEXT_ENCODER = "lora_te"
LORA_ADAPTER_NAME = "default"
def get_module_kohya_state_dict(
module: PeftModel, prefix: str, dtype: torch.dtype, adapter_name: str = LORA_ADAPTER_NAME
) -> Dict[str, torch.Tensor]:
kohya_ss_state_dict = {}
for peft_key, weight in get_peft_model_state_dict(module, adapter_name=adapter_name).items():
kohya_key = peft_key.replace("base_model.model", prefix)
kohya_key = kohya_key.replace("lora_A", "lora_down")
kohya_key = kohya_key.replace("lora_B", "lora_up")
kohya_key = kohya_key.replace(".", "_", kohya_key.count(".") - 2)
kohya_ss_state_dict[kohya_key] = weight.to(dtype)
# Set alpha parameter
if "lora_down" in kohya_key:
alpha_key = f'{kohya_key.split(".")[0]}.alpha'
kohya_ss_state_dict[alpha_key] = torch.tensor(module.peft_config[adapter_name].lora_alpha).to(dtype)
return kohya_ss_state_dict
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"--sd_checkpoint",
default=None,
type=str,
required=True,
help="Path to pretrained model or model identifier from huggingface.co/models.",
)
parser.add_argument(
"--sd_checkpoint_revision",
type=str,
default=None,
required=False,
help="Revision of pretrained model identifier from huggingface.co/models.",
)
parser.add_argument("--peft_lora_path", default=None, type=str, required=True, help="Path to peft trained LoRA")
parser.add_argument(
"--dump_path",
default=None,
type=str,
required=True,
help="Path to the output safetensors file for use with webui.",
)
parser.add_argument("--half", action="store_true", help="Save weights in half precision.")
args = parser.parse_args()
# Store kohya_ss state dict
kohya_ss_state_dict = {}
dtype = torch.float16 if args.half else torch.float32
# Load Text Encoder LoRA model
text_encoder_peft_lora_path = os.path.join(args.peft_lora_path, "text_encoder")
if os.path.exists(text_encoder_peft_lora_path):
text_encoder = CLIPTextModel.from_pretrained(
args.sd_checkpoint, subfolder="text_encoder", revision=args.sd_checkpoint_revision
)
text_encoder = PeftModel.from_pretrained(
text_encoder, text_encoder_peft_lora_path, adapter_name=LORA_ADAPTER_NAME
)
kohya_ss_state_dict.update(
get_module_kohya_state_dict(text_encoder, LORA_PREFIX_TEXT_ENCODER, dtype, LORA_ADAPTER_NAME)
)
# Load UNet LoRA model
unet_peft_lora_path = os.path.join(args.peft_lora_path, "unet")
if os.path.exists(unet_peft_lora_path):
unet = UNet2DConditionModel.from_pretrained(
args.sd_checkpoint, subfolder="unet", revision=args.sd_checkpoint_revision
)
unet = PeftModel.from_pretrained(unet, unet_peft_lora_path, adapter_name=LORA_ADAPTER_NAME)
kohya_ss_state_dict.update(get_module_kohya_state_dict(unet, LORA_PREFIX_UNET, dtype, LORA_ADAPTER_NAME))
# Save state dict
save_file(
kohya_ss_state_dict,
args.dump_path,
)
| peft/examples/lora_dreambooth/convert_peft_sd_lora_to_kohya_ss.py/0 | {
"file_path": "peft/examples/lora_dreambooth/convert_peft_sd_lora_to_kohya_ss.py",
"repo_id": "peft",
"token_count": 1639
} | 160 |
import argparse
import evaluate
import torch
from accelerate import Accelerator, DistributedDataParallelKwargs
from datasets import load_dataset
from torch.optim import AdamW
from torch.utils.data import DataLoader
from tqdm import tqdm
from transformers import AutoModelForSequenceClassification, AutoTokenizer, get_linear_schedule_with_warmup, set_seed
from peft import (
PrefixTuningConfig,
PromptEncoderConfig,
PromptTuningConfig,
get_peft_model,
)
from peft.utils.other import fsdp_auto_wrap_policy
def parse_args():
parser = argparse.ArgumentParser(description="PEFT a transformers model on a sequence classification task")
parser.add_argument(
"--num_virtual_tokens",
type=int,
default=20,
help="num_virtual_tokens if the number of virtual tokens used in prompt/prefix/P tuning.",
)
parser.add_argument(
"--encoder_hidden_size",
type=int,
default=128,
help="encoder_hidden_size if the encoder hidden size used in P tuninig/Prefix tuning.",
)
parser.add_argument(
"--model_name_or_path",
type=str,
help="Path to pretrained model or model identifier from huggingface.co/models.",
required=True,
)
parser.add_argument(
"--per_device_train_batch_size",
type=int,
default=8,
help="Batch size (per device) for the training dataloader.",
)
parser.add_argument(
"--per_device_eval_batch_size",
type=int,
default=8,
help="Batch size (per device) for the evaluation dataloader.",
)
parser.add_argument(
"--learning_rate",
type=float,
default=1e-3,
help="Initial learning rate (after the potential warmup period) to use.",
)
parser.add_argument("--num_train_epochs", type=int, default=3, help="Total number of training epochs to perform.")
parser.add_argument(
"--num_warmup_steps", type=int, default=0, help="Number of steps for the warmup in the lr scheduler."
)
parser.add_argument("--output_dir", type=str, default=None, help="Where to store the final model.")
parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
parser.add_argument(
"--peft_type",
type=str,
default="p_tuning",
help="The PEFT type to use.",
choices=["p_tuning", "prefix_tuning", "prompt_tuning"],
)
args = parser.parse_args()
assert args.output_dir is not None, "Need an `output_dir` to store the finetune model and verify."
return args
def main():
args = parse_args()
ddp_scaler = DistributedDataParallelKwargs(find_unused_parameters=True)
accelerator = Accelerator(kwargs_handlers=[ddp_scaler])
task = "mrpc"
# If passed along, set the training seed now.
if args.seed is not None:
set_seed(args.seed)
if args.peft_type == "p_tuning":
peft_config = PromptEncoderConfig(
task_type="SEQ_CLS",
num_virtual_tokens=args.num_virtual_tokens,
encoder_hidden_size=args.encoder_hidden_size,
)
elif args.peft_type == "prefix_tuning":
peft_config = PrefixTuningConfig(
task_type="SEQ_CLS",
num_virtual_tokens=args.num_virtual_tokens,
encoder_hidden_size=args.encoder_hidden_size,
)
else:
peft_config = PromptTuningConfig(task_type="SEQ_CLS", num_virtual_tokens=args.num_virtual_tokens)
tokenizer_kwargs = {}
if any(k in args.model_name_or_path for k in ("gpt", "opt", "bloom")):
tokenizer_kwargs["padding_side"] = "left"
else:
tokenizer_kwargs["padding_side"] = "right"
tokenizer = AutoTokenizer.from_pretrained(args.model_name_or_path, **tokenizer_kwargs)
if getattr(tokenizer, "pad_token_id") is None:
tokenizer.pad_token_id = tokenizer.eos_token_id
datasets = load_dataset("glue", task)
metric = evaluate.load("glue", task)
def tokenize_function(examples):
# max_length=None => use the model max length (it's actually the default)
outputs = tokenizer(examples["sentence1"], examples["sentence2"], truncation=True, max_length=None)
return outputs
def collate_fn(examples):
return tokenizer.pad(examples, padding="longest", return_tensors="pt")
with accelerator.main_process_first():
tokenized_datasets = datasets.map(
tokenize_function,
batched=True,
remove_columns=["idx", "sentence1", "sentence2"],
)
# We also rename the 'label' column to 'labels' which is the expected name for labels by the models of the
# transformers library
tokenized_datasets = tokenized_datasets.rename_column("label", "labels")
# Instantiate dataloaders.
train_dataloader = DataLoader(
tokenized_datasets["train"], shuffle=True, collate_fn=collate_fn, batch_size=args.per_device_train_batch_size
)
eval_dataloader = DataLoader(
tokenized_datasets["validation"],
shuffle=False,
collate_fn=collate_fn,
batch_size=args.per_device_eval_batch_size,
)
model = AutoModelForSequenceClassification.from_pretrained(args.model_name_or_path)
model = get_peft_model(model, peft_config)
model.print_trainable_parameters()
if getattr(accelerator.state, "fsdp_plugin", None) is not None:
accelerator.state.fsdp_plugin.auto_wrap_policy = fsdp_auto_wrap_policy(model)
model = accelerator.prepare(model)
optimizer = AdamW(params=model.parameters(), lr=args.learning_rate)
# Instantiate scheduler
lr_scheduler = get_linear_schedule_with_warmup(
optimizer=optimizer,
num_warmup_steps=args.num_warmup_steps,
num_training_steps=(len(train_dataloader) * args.num_train_epochs),
)
if getattr(accelerator.state, "fsdp_plugin", None) is not None:
train_dataloader, eval_dataloader, optimizer, lr_scheduler = accelerator.prepare(
train_dataloader, eval_dataloader, optimizer, lr_scheduler
)
else:
model, train_dataloader, eval_dataloader, optimizer, lr_scheduler = accelerator.prepare(
model, train_dataloader, eval_dataloader, optimizer, lr_scheduler
)
for epoch in range(args.num_train_epochs):
model.train()
for step, batch in enumerate(tqdm(train_dataloader)):
outputs = model(**batch)
loss = outputs.loss
accelerator.backward(loss)
optimizer.step()
lr_scheduler.step()
optimizer.zero_grad()
model.eval()
samples_seen = 0
for step, batch in enumerate(tqdm(eval_dataloader)):
with torch.no_grad():
outputs = model(**batch)
predictions = outputs.logits.argmax(dim=-1)
predictions, references = accelerator.gather((predictions, batch["labels"]))
# If we are in a multiprocess environment, the last batch has duplicates
if accelerator.num_processes > 1:
if step == len(eval_dataloader) - 1:
predictions = predictions[: len(eval_dataloader.dataset) - samples_seen]
references = references[: len(eval_dataloader.dataset) - samples_seen]
else:
samples_seen += references.shape[0]
metric.add_batch(
predictions=predictions,
references=references,
)
eval_metric = metric.compute()
accelerator.print(f"epoch {epoch}:", eval_metric)
accelerator.wait_for_everyone()
unwrapped_model = accelerator.unwrap_model(model)
unwrapped_model.save_pretrained(args.output_dir, state_dict=accelerator.get_state_dict(model))
if accelerator.is_main_process:
tokenizer.save_pretrained(args.output_dir)
if __name__ == "__main__":
main()
| peft/examples/sequence_classification/peft_no_lora_accelerate.py/0 | {
"file_path": "peft/examples/sequence_classification/peft_no_lora_accelerate.py",
"repo_id": "peft",
"token_count": 3361
} | 161 |
import inspect
from copy import deepcopy
from functools import update_wrapper
from types import MethodType
from .peft_model import PeftModel
def update_forward_signature(model: PeftModel) -> None:
"""
Args:
Updates the forward signature of the PeftModel to include parents class signature
model (`PeftModel`): Peft model to update the forward signature
Example:
```python
>>> from transformers import WhisperForConditionalGeneration
>>> from peft import get_peft_model, LoraConfig, update_forward_signature
>>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en")
>>> peft_config = LoraConfig(r=8, lora_alpha=32, lora_dropout=0.1, target_modules=["q_proj", "v_proj"])
>>> peft_model = get_peft_model(model, peft_config)
>>> update_forward_signature(peft_model)
```
"""
# Only update signature when the current forward signature only has *args and **kwargs
current_signature = inspect.signature(model.forward)
if (
len(current_signature.parameters) == 2
and "args" in current_signature.parameters
and "kwargs" in current_signature.parameters
):
forward = deepcopy(model.forward.__func__)
update_wrapper(
forward, type(model.get_base_model()).forward, assigned=("__doc__", "__name__", "__annotations__")
)
model.forward = MethodType(forward, model)
def update_generate_signature(model: PeftModel) -> None:
"""
Args:
Updates the generate signature of a PeftModel with overriding generate to include parents class signature
model (`PeftModel`): Peft model to update the generate signature
Example:
```python
>>> from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
>>> from peft import get_peft_model, LoraConfig, TaskType, update_generate_signature
>>> model_name_or_path = "bigscience/mt0-large"
>>> tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
>>> model = AutoModelForSeq2SeqLM.from_pretrained(model_name_or_path)
>>> peft_config = LoraConfig(
... task_type=TaskType.SEQ_2_SEQ_LM, inference_mode=False, r=8, lora_alpha=32, lora_dropout=0.1
... )
>>> peft_model = get_peft_model(model, peft_config)
>>> update_generate_signature(peft_model)
>>> help(peft_model.generate)
```
"""
if not hasattr(model, "generate"):
return
current_signature = inspect.signature(model.generate)
if (
len(current_signature.parameters) == 2
and "args" in current_signature.parameters
and "kwargs" in current_signature.parameters
) or (len(current_signature.parameters) == 1 and "kwargs" in current_signature.parameters):
generate = deepcopy(model.generate.__func__)
update_wrapper(
generate,
type(model.get_base_model()).generate,
assigned=("__doc__", "__name__", "__annotations__"),
)
model.generate = MethodType(generate, model)
def update_signature(model: PeftModel, method: str = "all") -> None:
"""
Args:
Updates the signature of a PeftModel include parents class signature for forward or generate method
model (`PeftModel`): Peft model to update generate or forward signature method (`str`): method to update
signature choose one of "forward", "generate", "all"
Example:
```python
>>> from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
>>> from peft import get_peft_model, LoraConfig, TaskType, update_signature
>>> model_name_or_path = "bigscience/mt0-large"
>>> tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
>>> model = AutoModelForSeq2SeqLM.from_pretrained(model_name_or_path)
>>> peft_config = LoraConfig(
... task_type=TaskType.SEQ_2_SEQ_LM, inference_mode=False, r=8, lora_alpha=32, lora_dropout=0.1
... )
>>> peft_model = get_peft_model(model, peft_config)
>>> update_signature(peft_model)
>>> help(peft_model.generate)
```
"""
if method == "forward":
update_forward_signature(model)
elif method == "generate":
update_generate_signature(model)
elif method == "all":
update_forward_signature(model)
update_generate_signature(model)
else:
raise ValueError(f"method {method} is not supported please choose one of ['forward', 'generate', 'all']")
| peft/src/peft/helpers.py/0 | {
"file_path": "peft/src/peft/helpers.py",
"repo_id": "peft",
"token_count": 1690
} | 162 |
# coding=utf-8
# Copyright 2023-present the HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from typing import Dict, List
import torch.nn as nn
from peft.utils import _freeze_adapter, _get_submodules
from .config import AdaptionPromptConfig, prepare_config
from .layer import AdaptedAttention
from .utils import is_adaption_prompt_trainable
class AdaptionPromptModel(nn.Module):
"""
Implements adaption prompts as described in https://arxiv.org/pdf/2303.16199.pdf.
The top L attention modules are replaced with AdaptedAttention modules that wrap the original ones, but insert
trainable prompts with gates (for zero init).
Notes on the multi-adapter pattern:
- We store the states of different adapters by keeping a dictionary of AdaptedAttention modules indexed by adapter
name.
- Every time we switch adapters, we remove the modules of the currently active adapter from the model, store them
in the dictionary, and replace them with the modules of the new adapter.
- To avoid duplicated and potentially inconsistent state, the currently active adapter is always removed from the
dictionary.
- Disabling the adapter would also result in the modules being removed from the model.
"""
def __init__(self, model, configs: Dict, adapter_name: str):
super().__init__()
self.model = model
# Store adapter configs by name.
self.peft_config: Dict[str, AdaptionPromptConfig] = {}
# Store lists of the parents of the affected attention modules by adapter name.
# We keep references to the parents so we can swap the adapters in-and-out of the model.
self._parents: Dict[str, List[nn.Module]] = {}
# Store lists of cached AdaptedAttention modules by name.
self._cached_adapters: Dict[str, List] = {}
# The name of the currently active adapter.
self._active_adapter = None
# Whether the adapter is enabled.
self._enabled = True
self.forward = self.model.forward
self.add_adapter(adapter_name, configs[adapter_name])
self._mark_only_adaption_prompts_as_trainable(self.model)
def add_adapter(self, adapter_name: str, config: AdaptionPromptConfig) -> None:
"""Add an adapter with the given name and config."""
config = prepare_config(config, self.model)
if adapter_name in self.peft_config:
raise ValueError(f"Adapter with name '{adapter_name}' already exists.")
parents = []
for name, _ in self.model.named_modules():
if name.endswith(config.target_modules):
par, _, _ = _get_submodules(self.model, name)
parents.append(par)
if len(parents) < config.adapter_layers:
raise ValueError(
f"Config specifies more adapter layers '{config.adapter_layers}'"
f" than the model has '{len(parents)}'."
)
# Note that if the target modules are not in Sequential, ModuleList, or
# some other PyTorch ordered container, the behavior is undefined as we
# assume here that the order of the modules is the same as the order of
# the transformer decoder layers.
parents = parents[-config.adapter_layers :]
self._parents[adapter_name] = parents
# It is only None during initialization.
# If it is disabled, we don't have to remove the modules.
if self._active_adapter is not None and self._enabled:
self._remove_adapted_attentions(self._active_adapter)
self._active_adapter = adapter_name
self.peft_config[adapter_name] = config
self._create_adapted_attentions(config, parents)
if not self._enabled:
self._remove_adapted_attentions(self._active_adapter)
if config.inference_mode:
_freeze_adapter(self.model, adapter_name)
def set_adapter(self, adapter_name: str) -> None:
"""Set the model to use the adapter with the given name."""
if self._active_adapter == adapter_name:
return
if adapter_name not in self.peft_config:
raise ValueError(f"Adapter with name '{adapter_name}' does not exist.")
if self._enabled:
self._remove_adapted_attentions(self._active_adapter)
self._set_adapted_attentions(adapter_name)
self._active_adapter = adapter_name
def enable_adapter_layers(self):
"""Enable adapter layers by swapping in cached AdaptedAttention modules."""
self._enabled = True
self._set_adapted_attentions(self._active_adapter)
def disable_adapter_layers(self):
"""Disable adapter layers by swapping out AdaptedAttention modules."""
self._enabled = False
self._remove_adapted_attentions(self._active_adapter)
def _create_adapted_attentions(self, config: AdaptionPromptConfig, parents: List[nn.Module]) -> None:
"""Wrap LlamaAttention modules with newly created AdaptedAttention modules."""
for par in parents:
attn = AdaptedAttention(
model_type=self.model.config.model_type,
adapter_len=config.adapter_len,
model=getattr(par, config.target_modules),
)
setattr(par, config.target_modules, attn)
def _set_adapted_attentions(self, adapter_name: str) -> None:
"""Replace LlamaAttention modules with cached AdaptedAttention modules."""
cached = self._cached_adapters[adapter_name]
del self._cached_adapters[adapter_name]
config = self.peft_config[adapter_name]
for i, par in enumerate(self._parents[adapter_name]):
setattr(par, config.target_modules, cached[i])
def _remove_adapted_attentions(self, adapter_name: str) -> None:
"""Remove AdaptedAttention modules from the model and store them in the cache."""
config = self.peft_config[adapter_name]
adapted_attentions = []
for par in self._parents[adapter_name]:
attn = getattr(par, config.target_modules)
adapted_attentions.append(attn)
setattr(par, config.target_modules, attn.model)
self._cached_adapters[adapter_name] = adapted_attentions
def _mark_only_adaption_prompts_as_trainable(self, model: nn.Module) -> None:
"""Freeze all parameters of the model except the adaption prompts."""
for n, p in model.named_parameters():
if not is_adaption_prompt_trainable(n):
p.requires_grad = False
def __getattr__(self, name: str):
"""Forward missing attributes to the wrapped module."""
try:
return super().__getattr__(name) # defer to nn.Module's logic
except AttributeError:
# This is necessary as e.g. causal models have various methods that we
# don't want to re-implement here.
return getattr(self.model, name)
| peft/src/peft/tuners/adaption_prompt/model.py/0 | {
"file_path": "peft/src/peft/tuners/adaption_prompt/model.py",
"repo_id": "peft",
"token_count": 2820
} | 163 |
# coding=utf-8
# Copyright 2023-present the HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import warnings
from typing import List, Optional
import bitsandbytes as bnb
import torch
from peft.import_utils import is_bnb_4bit_available, is_bnb_available
from peft.tuners.tuners_utils import BaseTunerLayer, check_adapters_to_merge
from peft.utils.other import transpose
from .layer import LoraLayer
if is_bnb_available():
class Linear8bitLt(torch.nn.Module, LoraLayer):
# Lora implemented in a dense layer
def __init__(
self,
base_layer: torch.nn.Module,
adapter_name: str,
r: int = 0,
lora_alpha: int = 1,
lora_dropout: float = 0.0,
init_lora_weights: bool = True,
use_rslora: bool = False,
**kwargs,
) -> None:
super().__init__()
LoraLayer.__init__(self, base_layer)
self._active_adapter = adapter_name
self.update_layer(adapter_name, r, lora_alpha, lora_dropout, init_lora_weights, use_rslora)
def merge(self, safe_merge: bool = False, adapter_names: Optional[List[str]] = None) -> None:
"""
Merge the active adapter weights into the base weights
Args:
safe_merge (`bool`, *optional*):
If True, the merge operation will be performed in a copy of the original weights and check for NaNs
before merging the weights. This is useful if you want to check if the merge operation will produce
NaNs. Defaults to `False`.
adapter_names (`List[str]`, *optional*):
The list of adapter names that should be merged. If None, all active adapters will be merged.
Defaults to `None`.
"""
adapter_names = check_adapters_to_merge(self, adapter_names)
if not adapter_names:
# no adapter to merge
return
for active_adapter in adapter_names:
if active_adapter not in self.lora_A.keys():
continue
warnings.warn(
"Merge lora module to 8-bit linear may get different generations due to rounding errors."
)
lora_data = self.get_delta_weight(active_adapter)
weight = self.get_base_layer().weight
state = self.get_base_layer().state
if state.SCB is None:
state.SCB = weight.SCB
# Dequantize the result of identity matrix and int8 weight because bitsandbytes does not support int8
# dequantization directly
im = torch.eye(weight.data.shape[-1]).contiguous().half().to(weight.device)
im, imt, SCim, SCimt, coo_tensorim = bnb.functional.double_quant(im)
im, Sim = bnb.functional.transform(im, "col32")
if state.CxB is None:
state.CxB, state.SB = bnb.functional.transform(weight.data, to_order=state.formatB)
out32, Sout32 = bnb.functional.igemmlt(im, state.CxB, Sim, state.SB)
output = bnb.functional.mm_dequant(out32, Sout32, SCim, state.SCB, bias=None).t()
w_data = output.to(lora_data.dtype).to(lora_data.device) + lora_data
if safe_merge and not torch.isfinite(w_data).all():
raise ValueError(
f"NaNs detected in the merged weights. The adapter {active_adapter} seems to be broken"
)
self.get_base_layer().weight = bnb.nn.Int8Params(
w_data.to("cpu"), requires_grad=False, has_fp16_weights=weight.has_fp16_weights
).to(weight.device)
state.reset_grads()
self.merged_adapters.append(active_adapter)
def unmerge(self) -> None:
"""
This method unmerges all merged adapter layers from the base weights.
"""
if not self.merged:
warnings.warn("Already unmerged. Nothing to do.")
return
while len(self.merged_adapters) > 0:
active_adapter = self.merged_adapters.pop()
if active_adapter not in self.lora_A.keys():
continue
warnings.warn(
"Unmerge lora module to 8-bit linear may get different generations due to rounding errors."
)
lora_data = self.get_delta_weight(active_adapter)
weight = self.get_base_layer().weight
state = self.get_base_layer().state
if state.SCB is None:
state.SCB = weight.SCB
im = torch.eye(weight.data.shape[-1]).contiguous().half().to(weight.device)
im, imt, SCim, SCimt, coo_tensorim = bnb.functional.double_quant(im)
im, Sim = bnb.functional.transform(im, "col32")
if state.CxB is None:
state.CxB, state.SB = bnb.functional.transform(weight.data, to_order=state.formatB)
out32, Sout32 = bnb.functional.igemmlt(im, state.CxB, Sim, state.SB)
output = bnb.functional.mm_dequant(out32, Sout32, SCim, state.SCB, bias=None).t()
w_data = output.to(lora_data.dtype).to(lora_data.device) - lora_data
self.get_base_layer().weight = bnb.nn.Int8Params(
w_data.to("cpu"), requires_grad=False, has_fp16_weights=weight.has_fp16_weights
).to(weight.device)
state.reset_grads()
def get_delta_weight(self, adapter):
return (
transpose(
self.lora_B[adapter].weight @ self.lora_A[adapter].weight,
False,
)
* self.scaling[adapter]
)
def forward(self, x: torch.Tensor, *args, **kwargs) -> torch.Tensor:
if self.disable_adapters:
if self.merged:
self.unmerge()
result = self.base_layer(x, *args, **kwargs)
elif self.merged:
result = self.base_layer(x, *args, **kwargs)
else:
result = self.base_layer(x, *args, **kwargs)
for active_adapter in self.active_adapters:
if active_adapter not in self.lora_A.keys():
continue
lora_A = self.lora_A[active_adapter]
lora_B = self.lora_B[active_adapter]
dropout = self.lora_dropout[active_adapter]
scaling = self.scaling[active_adapter]
requires_conversion = not torch.is_autocast_enabled()
if requires_conversion:
expected_dtype = result.dtype
compute_dtype = lora_A.weight.dtype
if x.dtype != compute_dtype:
x = x.to(compute_dtype)
output = lora_B(lora_A(dropout(x)))
if requires_conversion:
output = output.to(expected_dtype)
output = output * scaling
result = result + output
return result
def __repr__(self) -> str:
rep = super().__repr__()
return "lora." + rep
def dispatch_bnb_8bit(target: torch.nn.Module, adapter_name: str, **kwargs):
new_module = None
if isinstance(target, BaseTunerLayer):
target_base_layer = target.get_base_layer()
else:
target_base_layer = target
loaded_in_8bit = kwargs.get("loaded_in_8bit", False)
if loaded_in_8bit and isinstance(target_base_layer, bnb.nn.Linear8bitLt):
eightbit_kwargs = kwargs.copy()
eightbit_kwargs.update(
{
"has_fp16_weights": target.state.has_fp16_weights,
"memory_efficient_backward": target.state.memory_efficient_backward,
"threshold": target.state.threshold,
"index": target.index,
}
)
new_module = Linear8bitLt(target, adapter_name, **eightbit_kwargs)
return new_module
if is_bnb_4bit_available():
class Linear4bit(torch.nn.Module, LoraLayer):
# Lora implemented in a dense layer
def __init__(
self,
base_layer: torch.nn.Module,
adapter_name: str,
r: int = 0,
lora_alpha: int = 1,
lora_dropout: float = 0.0,
init_lora_weights: bool = True,
use_rslora: bool = False,
**kwargs,
) -> None:
super().__init__()
LoraLayer.__init__(self, base_layer)
self._active_adapter = adapter_name
self.update_layer(adapter_name, r, lora_alpha, lora_dropout, init_lora_weights, use_rslora)
def merge(self, safe_merge: bool = False, adapter_names: Optional[List[str]] = None) -> None:
"""
Merge the active adapter weights into the base weights
Args:
safe_merge (`bool`, *optional*):
If True, the merge operation will be performed in a copy of the original weights and check for NaNs
before merging the weights. This is useful if you want to check if the merge operation will produce
NaNs. Defaults to `False`.
adapter_names (`List[str]`, *optional*):
The list of adapter names that should be merged. If None, all active adapters will be merged.
Defaults to `None`.
"""
adapter_names = check_adapters_to_merge(self, adapter_names)
if not adapter_names:
# no adapter to merge
return
for active_adapter in adapter_names:
if active_adapter not in self.lora_A.keys():
continue
warnings.warn(
"Merge lora module to 4-bit linear may get different generations due to rounding errors."
)
# Refer to https://gist.github.com/ChrisHayduk/1a53463331f52dca205e55982baf9930
weight = self.get_base_layer().weight
kwargs = weight.__dict__
lora_data = self.get_delta_weight(active_adapter)
w_data = bnb.functional.dequantize_4bit(weight.data, weight.quant_state) + lora_data
if safe_merge and not torch.isfinite(w_data).all():
raise ValueError(
f"NaNs detected in the merged weights. The adapter {active_adapter} seems to be broken"
)
if "bnb_quantized" in kwargs:
kwargs["bnb_quantized"] = False
self.get_base_layer().weight = bnb.nn.Params4bit(w_data.to("cpu"), requires_grad=False, **kwargs).to(
weight.device
)
self.merged_adapters.append(active_adapter)
def unmerge(self) -> None:
"""
This method unmerges all merged adapter layers from the base weights.
"""
if not self.merged:
warnings.warn("Already unmerged. Nothing to do.")
return
while len(self.merged_adapters) > 0:
active_adapter = self.merged_adapters.pop()
if active_adapter not in self.lora_A.keys():
continue
warnings.warn(
"Unmerge lora module to 4-bit linear may get different generations due to rounding errors."
)
weight = self.get_base_layer().weight
kwargs = weight.__dict__
lora_data = self.get_delta_weight(active_adapter)
w_data = bnb.functional.dequantize_4bit(weight.data, weight.quant_state) - lora_data
if "bnb_quantized" in kwargs:
kwargs["bnb_quantized"] = False
self.get_base_layer().weight = bnb.nn.Params4bit(w_data.to("cpu"), requires_grad=False, **kwargs).to(
weight.device
)
def get_delta_weight(self, adapter):
return (
transpose(
self.lora_B[adapter].weight @ self.lora_A[adapter].weight,
False,
)
* self.scaling[adapter]
)
def forward(self, x: torch.Tensor, *args, **kwargs) -> torch.Tensor:
if self.disable_adapters:
if self.merged:
self.unmerge()
result = self.base_layer(x, *args, **kwargs)
elif self.merged:
result = self.base_layer(x, *args, **kwargs)
else:
result = self.base_layer(x, *args, **kwargs)
# As per Tim Dettmers, for 4bit, we need to defensively clone here.
# The reason is that in some cases, an error can occur that backprop
# does not work on a manipulated view. This issue may be solved with
# newer PyTorch versions but this would need extensive testing to be
# sure.
result = result.clone()
for active_adapter in self.active_adapters:
if active_adapter not in self.lora_A.keys():
continue
lora_A = self.lora_A[active_adapter]
lora_B = self.lora_B[active_adapter]
dropout = self.lora_dropout[active_adapter]
scaling = self.scaling[active_adapter]
requires_conversion = not torch.is_autocast_enabled()
if requires_conversion:
expected_dtype = result.dtype
x = x.to(lora_A.weight.dtype)
output = lora_B(lora_A(dropout(x)))
if requires_conversion:
output = output.to(expected_dtype)
output = output * scaling
result = result + output
return result
def __repr__(self) -> str:
rep = super().__repr__()
return "lora." + rep
def dispatch_bnb_4bit(target: torch.nn.Module, adapter_name: str, **kwargs):
new_module = None
if isinstance(target, BaseTunerLayer):
target_base_layer = target.get_base_layer()
else:
target_base_layer = target
loaded_in_4bit = kwargs.get("loaded_in_4bit", False)
if loaded_in_4bit and is_bnb_4bit_available() and isinstance(target_base_layer, bnb.nn.Linear4bit):
fourbit_kwargs = kwargs.copy()
fourbit_kwargs.update(
{
"compute_dtype": target_base_layer.compute_dtype,
"compress_statistics": target_base_layer.weight.compress_statistics,
"quant_type": target_base_layer.weight.quant_type,
}
)
new_module = Linear4bit(target, adapter_name, **fourbit_kwargs)
return new_module
| peft/src/peft/tuners/lora/bnb.py/0 | {
"file_path": "peft/src/peft/tuners/lora/bnb.py",
"repo_id": "peft",
"token_count": 8138
} | 164 |
# coding=utf-8
# Copyright 2023-present the HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import torch
# needed for prefix-tuning of bloom model
def bloom_model_postprocess_past_key_value(past_key_values):
past_key_values = torch.cat(past_key_values)
total_layers, batch_size, num_attention_heads, num_virtual_tokens, head_dim = past_key_values.shape
keys = past_key_values[: total_layers // 2]
keys = keys.transpose(2, 3).reshape(
total_layers // 2, batch_size * num_attention_heads, head_dim, num_virtual_tokens
)
values = past_key_values[total_layers // 2 :]
values = values.reshape(total_layers // 2, batch_size * num_attention_heads, num_virtual_tokens, head_dim)
return tuple(zip(keys, values))
# needed for prefix-tuning of StarCoder models
def starcoder_model_postprocess_past_key_value(past_key_values):
result = []
for k in past_key_values:
k = k[:, :, 0]
k = k.permute([1, 2, 0, 3])
k = k.reshape(*k.shape[:-2], -1)
result.append(k)
return tuple(result)
TRANSFORMERS_MODELS_TO_PREFIX_TUNING_POSTPROCESS_MAPPING = {
"bloom": bloom_model_postprocess_past_key_value,
"gpt_bigcode": starcoder_model_postprocess_past_key_value,
}
TRANSFORMERS_MODELS_TO_LORA_TARGET_MODULES_MAPPING = {
"t5": ["q", "v"],
"mt5": ["q", "v"],
"bart": ["q_proj", "v_proj"],
"gpt2": ["c_attn"],
"bloom": ["query_key_value"],
"blip-2": ["q", "v", "q_proj", "v_proj"],
"opt": ["q_proj", "v_proj"],
"gptj": ["q_proj", "v_proj"],
"gpt_neox": ["query_key_value"],
"gpt_neo": ["q_proj", "v_proj"],
"bert": ["query", "value"],
"roberta": ["query", "value"],
"xlm-roberta": ["query", "value"],
"electra": ["query", "value"],
"deberta-v2": ["query_proj", "value_proj"],
"deberta": ["in_proj"],
"layoutlm": ["query", "value"],
"llama": ["q_proj", "v_proj"],
"chatglm": ["query_key_value"],
"gpt_bigcode": ["c_attn"],
"mpt": ["Wqkv"],
"RefinedWebModel": ["query_key_value"],
"RefinedWeb": ["query_key_value"],
"falcon": ["query_key_value"],
"btlm": ["c_proj", "c_attn"],
"codegen": ["qkv_proj"],
"mistral": ["q_proj", "v_proj"],
"mixtral": ["q_proj", "v_proj"],
"stablelm": ["q_proj", "v_proj"],
"phi": ["q_proj", "v_proj", "fc1", "fc2"],
}
TRANSFORMERS_MODELS_TO_IA3_TARGET_MODULES_MAPPING = {
"t5": ["k", "v", "wo"],
"mt5": ["k", "v", "wi_1"],
"gpt2": ["c_attn", "mlp.c_proj"],
"bloom": ["query_key_value", "mlp.dense_4h_to_h"],
"roberta": ["key", "value", "output.dense"],
"opt": ["q_proj", "k_proj", "fc2"],
"gptj": ["q_proj", "v_proj", "fc_out"],
"gpt_neox": ["query_key_value", "dense_4h_to_h"],
"gpt_neo": ["q_proj", "v_proj", "c_proj"],
"bart": ["q_proj", "v_proj", "fc2"],
"gpt_bigcode": ["c_attn", "mlp.c_proj"],
"llama": ["k_proj", "v_proj", "down_proj"],
"mistral": ["k_proj", "v_proj", "down_proj"],
"bert": ["key", "value", "output.dense"],
"deberta-v2": ["key_proj", "value_proj", "output.dense"],
"deberta": ["in_proj", "output.dense"],
"RefinedWebModel": ["query_key_value", "dense_4h_to_h"],
"RefinedWeb": ["query_key_value", "dense_4h_to_h"],
"falcon": ["query_key_value", "dense_4h_to_h"],
"phi": ["q_proj", "v_proj", "fc2"],
}
TRANSFORMERS_MODELS_TO_IA3_FEEDFORWARD_MODULES_MAPPING = {
"t5": ["wo"],
"mt5": [],
"gpt2": ["mlp.c_proj"],
"bloom": ["mlp.dense_4h_to_h"],
"roberta": ["output.dense"],
"opt": ["fc2"],
"gptj": ["fc_out"],
"gpt_neox": ["dense_4h_to_h"],
"gpt_neo": ["c_proj"],
"bart": ["fc2"],
"gpt_bigcode": ["mlp.c_proj"],
"llama": ["down_proj"],
"mistral": ["down_proj"],
"bert": ["output.dense"],
"deberta-v2": ["output.dense"],
"deberta": ["output.dense"],
"RefinedWeb": ["dense_4h_to_h"],
"RefinedWebModel": ["dense_4h_to_h"],
"falcon": ["dense_4h_to_h"],
"phi": ["fc2"],
}
TRANSFORMERS_MODELS_TO_ADALORA_TARGET_MODULES_MAPPING = {
"t5": ["q", "k", "v", "o", "wi", "wo"],
"mt5": ["q", "k", "v", "o", "wi_0", "wi_1", "wo"],
"bart": ["q_proj", "k_proj", "v_proj", "out_proj", "fc1", "fc2"],
"gpt2": ["c_attn"],
"bloom": ["query_key_value"],
"opt": ["q_proj", "k_proj", "v_proj", "out_proj", "fc1", "fc2"],
"gptj": ["q_proj", "v_proj"],
"gpt_neox": ["query_key_value"],
"gpt_neo": ["q_proj", "v_proj"],
"llama": ["q_proj", "v_proj"],
"bert": ["query", "value"],
"roberta": ["query", "key", "value", "dense"],
# "xlm-roberta": ["query", "value"],
# "electra": ["query", "value"],
"deberta-v2": ["query_proj", "key_proj", "value_proj", "dense"],
"gpt_bigcode": ["c_attn"],
"deberta": ["in_proj"],
# "layoutlm": ["query", "value"],
}
WEIGHTS_NAME = "adapter_model.bin"
SAFETENSORS_WEIGHTS_NAME = "adapter_model.safetensors"
CONFIG_NAME = "adapter_config.json"
EMBEDDING_LAYER_NAMES = ["embed_tokens", "lm_head"]
INCLUDE_LINEAR_LAYERS_SHORTHAND = "all-linear"
TOKENIZER_CONFIG_NAME = "tokenizer_config.json"
| peft/src/peft/utils/constants.py/0 | {
"file_path": "peft/src/peft/utils/constants.py",
"repo_id": "peft",
"token_count": 2630
} | 165 |
# coding=utf-8
# Copyright 2023-present the HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import gc
import os
import tempfile
import unittest
from dataclasses import dataclass
from typing import Any, Dict, List, Union
import pytest
import torch
from accelerate import infer_auto_device_map
from accelerate.test_utils.testing import run_command
from accelerate.utils import patch_environment
from datasets import Audio, DatasetDict, load_dataset
from parameterized import parameterized
from transformers import (
AutoModelForCausalLM,
AutoModelForSeq2SeqLM,
AutoTokenizer,
DataCollatorForLanguageModeling,
Seq2SeqTrainer,
Seq2SeqTrainingArguments,
Trainer,
TrainingArguments,
WhisperFeatureExtractor,
WhisperForConditionalGeneration,
WhisperProcessor,
WhisperTokenizer,
)
from peft import (
AdaLoraConfig,
LoftQConfig,
LoraConfig,
PeftModel,
TaskType,
get_peft_model,
prepare_model_for_int8_training,
prepare_model_for_kbit_training,
)
from peft.utils import SAFETENSORS_WEIGHTS_NAME
from .testing_utils import (
require_auto_gptq,
require_bitsandbytes,
require_optimum,
require_torch_gpu,
require_torch_multi_gpu,
)
# A full testing suite that tests all the necessary features on GPU. The tests should
# rely on the example scripts to test the features.
@dataclass
class DataCollatorSpeechSeq2SeqWithPadding:
r"""
Directly copied from:
https://github.com/huggingface/peft/blob/main/examples/int8_training/peft_bnb_whisper_large_v2_training.ipynb
"""
processor: Any
def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]:
# split inputs and labels since they have to be of different lengths and need different padding methods
# first treat the audio inputs by simply returning torch tensors
input_features = [{"input_features": feature["input_features"]} for feature in features]
batch = self.processor.feature_extractor.pad(input_features, return_tensors="pt")
# get the tokenized label sequences
label_features = [{"input_ids": feature["labels"]} for feature in features]
# pad the labels to max length
labels_batch = self.processor.tokenizer.pad(label_features, return_tensors="pt")
# replace padding with -100 to ignore loss correctly
labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)
# if bos token is appended in previous tokenization step,
# cut bos token here as it's append later anyways
if (labels[:, 0] == self.processor.tokenizer.bos_token_id).all().cpu().item():
labels = labels[:, 1:]
batch["labels"] = labels
return batch
@require_torch_gpu
@require_bitsandbytes
class PeftBnbGPUExampleTests(unittest.TestCase):
r"""
A single GPU int8 + fp4 test suite, this will test if training fits correctly on a single GPU device (1x NVIDIA T4
16GB) using bitsandbytes.
The tests are the following:
- Seq2Seq model training based on:
https://github.com/huggingface/peft/blob/main/examples/int8_training/Finetune_flan_t5_large_bnb_peft.ipynb
- Causal LM model training based on:
https://github.com/huggingface/peft/blob/main/examples/int8_training/Finetune_opt_bnb_peft.ipynb
- Audio model training based on:
https://github.com/huggingface/peft/blob/main/examples/int8_training/peft_bnb_whisper_large_v2_training.ipynb
"""
def setUp(self):
self.seq2seq_model_id = "google/flan-t5-base"
self.causal_lm_model_id = "facebook/opt-6.7b"
self.tokenizer = AutoTokenizer.from_pretrained(self.causal_lm_model_id)
self.audio_model_id = "openai/whisper-large"
def tearDown(self):
r"""
Efficient mechanism to free GPU memory after each test. Based on
https://github.com/huggingface/transformers/issues/21094
"""
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
gc.collect()
def _check_inference_finite(self, model, batch):
# try inference without Trainer class
training = model.training
model.eval()
output = model(**batch.to(model.device))
self.assertTrue(torch.isfinite(output.logits).all())
model.train(training)
@pytest.mark.single_gpu_tests
def test_causal_lm_training(self):
r"""
Test the CausalLM training on a single GPU device. This test is a converted version of
https://github.com/huggingface/peft/blob/main/examples/int8_training/Finetune_opt_bnb_peft.ipynb where we train
`opt-6.7b` on `english_quotes` dataset in few steps. The test would simply fail if the adapters are not set
correctly.
"""
with tempfile.TemporaryDirectory() as tmp_dir:
model = AutoModelForCausalLM.from_pretrained(
self.causal_lm_model_id,
load_in_8bit=True,
device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained(self.causal_lm_model_id)
model = prepare_model_for_int8_training(model)
config = LoraConfig(
r=16,
lora_alpha=32,
target_modules=["q_proj", "v_proj"],
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, config)
data = load_dataset("ybelkada/english_quotes_copy")
data = data.map(lambda samples: tokenizer(samples["quote"]), batched=True)
trainer = Trainer(
model=model,
train_dataset=data["train"],
args=TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
warmup_steps=2,
max_steps=3,
learning_rate=2e-4,
fp16=True,
logging_steps=1,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.single_gpu_tests
def test_causal_lm_training_4bit(self):
r"""
Test the CausalLM training on a single GPU device. This test is a converted version of
https://github.com/huggingface/peft/blob/main/examples/int8_training/Finetune_opt_bnb_peft.ipynb where we train
`opt-6.7b` on `english_quotes` dataset in few steps using 4bit base model. The test would simply fail if the
adapters are not set correctly.
"""
with tempfile.TemporaryDirectory() as tmp_dir:
model = AutoModelForCausalLM.from_pretrained(
self.causal_lm_model_id,
load_in_4bit=True,
device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained(self.causal_lm_model_id)
model = prepare_model_for_kbit_training(model)
config = LoraConfig(
r=16,
lora_alpha=32,
target_modules=["q_proj", "v_proj"],
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, config)
data = load_dataset("ybelkada/english_quotes_copy")
data = data.map(lambda samples: tokenizer(samples["quote"]), batched=True)
trainer = Trainer(
model=model,
train_dataset=data["train"],
args=TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
warmup_steps=2,
max_steps=3,
learning_rate=2e-4,
fp16=True,
logging_steps=1,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.multi_gpu_tests
def test_causal_lm_training_multi_gpu_4bit(self):
r"""
Test the CausalLM training on a multi-GPU device with 4bit base model. The test would simply fail if the
adapters are not set correctly.
"""
with tempfile.TemporaryDirectory() as tmp_dir:
model = AutoModelForCausalLM.from_pretrained(
self.causal_lm_model_id,
device_map="auto",
load_in_4bit=True,
)
self.assertEqual(set(model.hf_device_map.values()), set(range(torch.cuda.device_count())))
model = prepare_model_for_kbit_training(model)
setattr(model, "model_parallel", True)
setattr(model, "is_parallelizable", True)
config = LoraConfig(
r=16,
lora_alpha=32,
target_modules=["q_proj", "v_proj"],
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, config)
data = load_dataset("Abirate/english_quotes")
data = data.map(lambda samples: self.tokenizer(samples["quote"]), batched=True)
trainer = Trainer(
model=model,
train_dataset=data["train"],
args=TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
warmup_steps=2,
max_steps=3,
learning_rate=2e-4,
fp16=True,
logging_steps=1,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(self.tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.single_gpu_tests
@require_torch_gpu
def test_4bit_adalora_causalLM(self):
r"""
Tests the 4bit training with adalora
"""
model_id = "facebook/opt-350m"
# for >3 GPUs, might need: device_map={"": "cuda:0"}
model = AutoModelForCausalLM.from_pretrained(model_id, load_in_4bit=True)
tokenizer = AutoTokenizer.from_pretrained(model_id)
model.gradient_checkpointing_enable()
model = prepare_model_for_kbit_training(model)
peft_config = AdaLoraConfig(
init_r=6,
target_r=4,
tinit=50,
tfinal=100,
deltaT=5,
beta1=0.3,
beta2=0.3,
orth_reg_weight=0.2,
lora_alpha=32,
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, peft_config)
data = load_dataset("ybelkada/english_quotes_copy")
data = data.map(lambda samples: tokenizer(samples["quote"]), batched=True)
batch = tokenizer(data["train"][:3]["quote"], return_tensors="pt", padding=True)
self._check_inference_finite(model, batch)
with tempfile.TemporaryDirectory() as tmp_dir:
trainer = Trainer(
model=model,
train_dataset=data["train"],
args=TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
warmup_steps=2,
max_steps=3,
learning_rate=2e-4,
fp16=True,
logging_steps=1,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.single_gpu_tests
@require_torch_gpu
def test_8bit_adalora_causalLM(self):
r"""
Tests the 8bit training with adalora
"""
model_id = "facebook/opt-350m"
model = AutoModelForCausalLM.from_pretrained(model_id, load_in_8bit=True)
tokenizer = AutoTokenizer.from_pretrained(model_id)
model.gradient_checkpointing_enable()
model = prepare_model_for_kbit_training(model)
peft_config = AdaLoraConfig(
init_r=6,
target_r=4,
tinit=50,
tfinal=100,
deltaT=5,
beta1=0.3,
beta2=0.3,
orth_reg_weight=0.2,
lora_alpha=32,
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, peft_config)
data = load_dataset("ybelkada/english_quotes_copy")
data = data.map(lambda samples: tokenizer(samples["quote"]), batched=True)
batch = tokenizer(data["train"][:3]["quote"], return_tensors="pt", padding=True)
self._check_inference_finite(model, batch)
with tempfile.TemporaryDirectory() as tmp_dir:
trainer = Trainer(
model=model,
train_dataset=data["train"],
args=TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
warmup_steps=2,
max_steps=3,
learning_rate=2e-4,
fp16=True,
logging_steps=1,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.multi_gpu_tests
@require_torch_multi_gpu
def test_causal_lm_training_multi_gpu(self):
r"""
Test the CausalLM training on a multi-GPU device. This test is a converted version of
https://github.com/huggingface/peft/blob/main/examples/int8_training/Finetune_opt_bnb_peft.ipynb where we train
`opt-6.7b` on `english_quotes` dataset in few steps. The test would simply fail if the adapters are not set
correctly.
"""
with tempfile.TemporaryDirectory() as tmp_dir:
model = AutoModelForCausalLM.from_pretrained(
self.causal_lm_model_id,
load_in_8bit=True,
device_map="auto",
)
self.assertEqual(set(model.hf_device_map.values()), set(range(torch.cuda.device_count())))
tokenizer = AutoTokenizer.from_pretrained(self.causal_lm_model_id)
model = prepare_model_for_int8_training(model)
setattr(model, "model_parallel", True)
setattr(model, "is_parallelizable", True)
config = LoraConfig(
r=16,
lora_alpha=32,
target_modules=["q_proj", "v_proj"],
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, config)
data = load_dataset("Abirate/english_quotes")
data = data.map(lambda samples: tokenizer(samples["quote"]), batched=True)
trainer = Trainer(
model=model,
train_dataset=data["train"],
args=TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
warmup_steps=2,
max_steps=3,
learning_rate=2e-4,
fp16=True,
logging_steps=1,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.single_gpu_tests
def test_seq2seq_lm_training_single_gpu(self):
r"""
Test the Seq2SeqLM training on a single GPU device. This test is a converted version of
https://github.com/huggingface/peft/blob/main/examples/int8_training/Finetune_opt_bnb_peft.ipynb where we train
`flan-large` on `english_quotes` dataset in few steps. The test would simply fail if the adapters are not set
correctly.
"""
with tempfile.TemporaryDirectory() as tmp_dir:
model = AutoModelForSeq2SeqLM.from_pretrained(
self.seq2seq_model_id,
load_in_8bit=True,
device_map={"": 0},
)
self.assertEqual(set(model.hf_device_map.values()), {0})
tokenizer = AutoTokenizer.from_pretrained(self.seq2seq_model_id)
model = prepare_model_for_int8_training(model)
config = LoraConfig(
r=16,
lora_alpha=32,
target_modules=["q", "v"],
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, config)
data = load_dataset("ybelkada/english_quotes_copy")
data = data.map(lambda samples: tokenizer(samples["quote"]), batched=True)
trainer = Trainer(
model=model,
train_dataset=data["train"],
args=TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
warmup_steps=2,
max_steps=3,
learning_rate=2e-4,
fp16=True,
logging_steps=1,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.multi_gpu_tests
@require_torch_multi_gpu
def test_seq2seq_lm_training_multi_gpu(self):
r"""
Test the Seq2SeqLM training on a multi-GPU device. This test is a converted version of
https://github.com/huggingface/peft/blob/main/examples/int8_training/Finetune_opt_bnb_peft.ipynb where we train
`flan-large` on `english_quotes` dataset in few steps. The test would simply fail if the adapters are not set
correctly.
"""
with tempfile.TemporaryDirectory() as tmp_dir:
model = AutoModelForSeq2SeqLM.from_pretrained(
self.seq2seq_model_id,
load_in_8bit=True,
device_map="balanced",
)
self.assertEqual(set(model.hf_device_map.values()), set(range(torch.cuda.device_count())))
tokenizer = AutoTokenizer.from_pretrained(self.seq2seq_model_id)
model = prepare_model_for_int8_training(model)
config = LoraConfig(
r=16,
lora_alpha=32,
target_modules=["q", "v"],
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, config)
data = load_dataset("ybelkada/english_quotes_copy")
data = data.map(lambda samples: tokenizer(samples["quote"]), batched=True)
trainer = Trainer(
model=model,
train_dataset=data["train"],
args=TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
warmup_steps=2,
max_steps=3,
learning_rate=2e-4,
fp16=True,
logging_steps=1,
output_dir="outputs",
),
data_collator=DataCollatorForLanguageModeling(tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.single_gpu_tests
def test_audio_model_training(self):
r"""
Test the audio model training on a single GPU device. This test is a converted version of
https://github.com/huggingface/peft/blob/main/examples/int8_training/peft_bnb_whisper_large_v2_training.ipynb
"""
with tempfile.TemporaryDirectory() as tmp_dir:
dataset_name = "ybelkada/common_voice_mr_11_0_copy"
task = "transcribe"
language = "Marathi"
common_voice = DatasetDict()
common_voice["train"] = load_dataset(dataset_name, split="train+validation")
common_voice = common_voice.remove_columns(
["accent", "age", "client_id", "down_votes", "gender", "locale", "path", "segment", "up_votes"]
)
feature_extractor = WhisperFeatureExtractor.from_pretrained(self.audio_model_id)
tokenizer = WhisperTokenizer.from_pretrained(self.audio_model_id, language=language, task=task)
processor = WhisperProcessor.from_pretrained(self.audio_model_id, language=language, task=task)
common_voice = common_voice.cast_column("audio", Audio(sampling_rate=16000))
def prepare_dataset(batch):
# load and resample audio data from 48 to 16kHz
audio = batch["audio"]
# compute log-Mel input features from input audio array
batch["input_features"] = feature_extractor(
audio["array"], sampling_rate=audio["sampling_rate"]
).input_features[0]
# encode target text to label ids
batch["labels"] = tokenizer(batch["sentence"]).input_ids
return batch
common_voice = common_voice.map(
prepare_dataset, remove_columns=common_voice.column_names["train"], num_proc=2
)
data_collator = DataCollatorSpeechSeq2SeqWithPadding(processor=processor)
model = WhisperForConditionalGeneration.from_pretrained(
self.audio_model_id, load_in_8bit=True, device_map="auto"
)
model.config.forced_decoder_ids = None
model.config.suppress_tokens = []
model = prepare_model_for_int8_training(model)
# as Whisper model uses Conv layer in encoder, checkpointing disables grad computation
# to avoid this, make the inputs trainable
def make_inputs_require_grad(module, input, output):
output.requires_grad_(True)
model.model.encoder.conv1.register_forward_hook(make_inputs_require_grad)
config = LoraConfig(
r=32, lora_alpha=64, target_modules=["q_proj", "v_proj"], lora_dropout=0.05, bias="none"
)
model = get_peft_model(model, config)
model.print_trainable_parameters()
training_args = Seq2SeqTrainingArguments(
output_dir=tmp_dir, # change to a repo name of your choice
per_device_train_batch_size=8,
gradient_accumulation_steps=1, # increase by 2x for every 2x decrease in batch size
learning_rate=1e-3,
warmup_steps=2,
max_steps=3,
fp16=True,
per_device_eval_batch_size=8,
generation_max_length=128,
logging_steps=25,
remove_unused_columns=False, # required as the PeftModel forward doesn't have the signature of the wrapped model's forward
label_names=["labels"], # same reason as above
)
trainer = Seq2SeqTrainer(
args=training_args,
model=model,
train_dataset=common_voice["train"],
data_collator=data_collator,
tokenizer=processor.feature_extractor,
)
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.single_gpu_tests
def test_4bit_non_default_adapter_name(self):
# See PR 1294
config = LoraConfig(
r=16,
target_modules=["q_proj", "v_proj"],
bias="none",
task_type="CAUSAL_LM",
)
# default adapter name
model = AutoModelForCausalLM.from_pretrained(
"facebook/opt-125m",
device_map="auto",
load_in_4bit=True,
)
model = prepare_model_for_kbit_training(model)
model = get_peft_model(model, config)
n_trainable_default, n_total_default = model.get_nb_trainable_parameters()
# other adapter name
model = AutoModelForCausalLM.from_pretrained(
"facebook/opt-125m",
device_map="auto",
load_in_4bit=True,
)
model = prepare_model_for_kbit_training(model)
model = get_peft_model(model, config, adapter_name="other")
n_trainable_other, n_total_other = model.get_nb_trainable_parameters()
self.assertGreater(n_trainable_other, 0) # sanity check
self.assertEqual(n_trainable_default, n_trainable_other)
self.assertEqual(n_total_default, n_total_other)
@pytest.mark.single_gpu_tests
def test_8bit_non_default_adapter_name(self):
# See PR 1294
config = LoraConfig(
r=16,
target_modules=["q_proj", "v_proj"],
bias="none",
task_type="CAUSAL_LM",
)
# default adapter name
model = AutoModelForCausalLM.from_pretrained(
"facebook/opt-125m",
device_map="auto",
load_in_8bit=True,
)
model = prepare_model_for_kbit_training(model)
model = get_peft_model(model, config)
n_trainable_default, n_total_default = model.get_nb_trainable_parameters()
# other adapter name
model = AutoModelForCausalLM.from_pretrained(
"facebook/opt-125m",
device_map="auto",
load_in_8bit=True,
)
model = prepare_model_for_kbit_training(model)
model = get_peft_model(model, config, adapter_name="other")
n_trainable_other, n_total_other = model.get_nb_trainable_parameters()
self.assertGreater(n_trainable_other, 0) # sanity check
self.assertEqual(n_trainable_default, n_trainable_other)
self.assertEqual(n_total_default, n_total_other)
@require_torch_gpu
@require_auto_gptq
@require_optimum
class PeftGPTQGPUTests(unittest.TestCase):
r"""
GPTQ + peft tests
"""
def setUp(self):
from transformers import GPTQConfig
self.causal_lm_model_id = "marcsun13/opt-350m-gptq-4bit"
# TODO : check if it works for Exllamav2 kernels
self.quantization_config = GPTQConfig(bits=4, use_exllama=False)
self.tokenizer = AutoTokenizer.from_pretrained(self.causal_lm_model_id)
def tearDown(self):
r"""
Efficient mechanism to free GPU memory after each test. Based on
https://github.com/huggingface/transformers/issues/21094
"""
gc.collect()
torch.cuda.empty_cache()
def _check_inference_finite(self, model, batch):
# try inference without Trainer class
training = model.training
model.eval()
output = model(**batch.to(model.device))
self.assertTrue(torch.isfinite(output.logits).all())
model.train(training)
@pytest.mark.single_gpu_tests
def test_causal_lm_training(self):
r"""
Test the CausalLM training on a single GPU device. The test would simply fail if the adapters are not set
correctly.
"""
with tempfile.TemporaryDirectory() as tmp_dir:
model = AutoModelForCausalLM.from_pretrained(
self.causal_lm_model_id,
torch_dtype=torch.float16,
device_map="auto",
quantization_config=self.quantization_config,
)
model = prepare_model_for_kbit_training(model)
config = LoraConfig(
r=16,
lora_alpha=32,
target_modules=["q_proj", "v_proj"],
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, config)
data = load_dataset("ybelkada/english_quotes_copy")
data = data.map(lambda samples: self.tokenizer(samples["quote"]), batched=True)
trainer = Trainer(
model=model,
train_dataset=data["train"],
args=TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
warmup_steps=2,
max_steps=3,
learning_rate=2e-4,
fp16=True,
logging_steps=1,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(self.tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.single_gpu_tests
def test_adalora_causalLM(self):
r"""
Tests the gptq training with adalora
"""
model = AutoModelForCausalLM.from_pretrained(
self.causal_lm_model_id,
torch_dtype=torch.float16,
device_map="auto",
quantization_config=self.quantization_config,
)
tokenizer = AutoTokenizer.from_pretrained(self.causal_lm_model_id)
model = prepare_model_for_kbit_training(model)
peft_config = AdaLoraConfig(
init_r=6,
target_r=4,
tinit=50,
tfinal=100,
deltaT=5,
beta1=0.3,
beta2=0.3,
orth_reg_weight=0.2,
lora_alpha=32,
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, peft_config)
data = load_dataset("ybelkada/english_quotes_copy")
data = data.map(lambda samples: self.tokenizer(samples["quote"]), batched=True)
batch = tokenizer(data["train"][:3]["quote"], return_tensors="pt", padding=True)
self._check_inference_finite(model, batch)
with tempfile.TemporaryDirectory() as tmp_dir:
trainer = Trainer(
model=model,
train_dataset=data["train"],
args=TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
warmup_steps=2,
max_steps=3,
learning_rate=2e-4,
fp16=True,
logging_steps=1,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(self.tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.multi_gpu_tests
@require_torch_multi_gpu
def test_causal_lm_training_multi_gpu(self):
r"""
Test the CausalLM training on a multi-GPU device. The test would simply fail if the adapters are not set
correctly.
"""
with tempfile.TemporaryDirectory() as tmp_dir:
model = AutoModelForCausalLM.from_pretrained(
self.causal_lm_model_id,
torch_dtype=torch.float16,
device_map="auto",
quantization_config=self.quantization_config,
)
self.assertEqual(set(model.hf_device_map.values()), set(range(torch.cuda.device_count())))
model = prepare_model_for_kbit_training(model)
setattr(model, "model_parallel", True)
setattr(model, "is_parallelizable", True)
config = LoraConfig(
r=16,
lora_alpha=32,
target_modules=["q_proj", "v_proj"],
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, config)
data = load_dataset("Abirate/english_quotes")
data = data.map(lambda samples: self.tokenizer(samples["quote"]), batched=True)
trainer = Trainer(
model=model,
train_dataset=data["train"],
args=TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
warmup_steps=2,
max_steps=3,
learning_rate=2e-4,
fp16=True,
logging_steps=1,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(self.tokenizer, mlm=False),
)
model.config.use_cache = False
trainer.train()
model.cpu().save_pretrained(tmp_dir)
self.assertTrue("adapter_config.json" in os.listdir(tmp_dir))
self.assertTrue(SAFETENSORS_WEIGHTS_NAME in os.listdir(tmp_dir))
# assert loss is not None
self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
@pytest.mark.single_gpu_tests
def test_non_default_adapter_name(self):
# See issue 1346
config = LoraConfig(
r=16,
target_modules=["q_proj", "v_proj"],
task_type="CAUSAL_LM",
)
# default adapter name
model = AutoModelForCausalLM.from_pretrained(
self.causal_lm_model_id,
torch_dtype=torch.float16,
device_map="auto",
quantization_config=self.quantization_config,
)
model = prepare_model_for_kbit_training(model)
model = get_peft_model(model, config)
n_trainable_default, n_total_default = model.get_nb_trainable_parameters()
# other adapter name
model = AutoModelForCausalLM.from_pretrained(
self.causal_lm_model_id,
torch_dtype=torch.float16,
device_map="auto",
quantization_config=self.quantization_config,
)
model = prepare_model_for_kbit_training(model)
model = get_peft_model(model, config, adapter_name="other")
n_trainable_other, n_total_other = model.get_nb_trainable_parameters()
self.assertGreater(n_trainable_other, 0) # sanity check
self.assertEqual(n_trainable_default, n_trainable_other)
self.assertEqual(n_total_default, n_total_other)
@require_torch_gpu
class OffloadSaveTests(unittest.TestCase):
def setUp(self):
self.causal_lm_model_id = "gpt2"
def tearDown(self):
r"""
Efficient mechanism to free GPU memory after each test. Based on
https://github.com/huggingface/transformers/issues/21094
"""
gc.collect()
torch.cuda.empty_cache()
@pytest.mark.single_gpu_tests
@require_torch_gpu
def test_offload_merge(self):
r"""
Test merging, unmerging, and unloading of a model with CPU- offloaded modules.
"""
torch.manual_seed(0)
model = AutoModelForCausalLM.from_pretrained(self.causal_lm_model_id)
tokenizer = AutoTokenizer.from_pretrained(self.causal_lm_model_id)
# TODO: add disk offload once PeftModel.from_pretrained supports
memory_limits = {0: "0.4GIB", "cpu": "5GIB"}
# offloads around half of all transformer modules
device_map = infer_auto_device_map(model, max_memory=memory_limits)
self.assertTrue(0 in device_map.values())
self.assertTrue("cpu" in device_map.values())
config = LoraConfig(task_type="CAUSAL_LM", init_lora_weights=False, target_modules=["c_attn"])
model = get_peft_model(model, config)
with tempfile.TemporaryDirectory() as tmp_dir:
model.save_pretrained(tmp_dir)
# load the model with device_map
model = AutoModelForCausalLM.from_pretrained(self.causal_lm_model_id, device_map=device_map).eval()
self.assertTrue(len({p.device for p in model.parameters()}) == 2)
model = PeftModel.from_pretrained(model, tmp_dir, max_memory=memory_limits)
input_tokens = tokenizer.encode("Four score and seven years ago", return_tensors="pt")
model.eval()
# test peft model adapter merge
pre_merge_olayer = model(input_tokens)[0]
model.merge_adapter()
post_merge_olayer = model(input_tokens)[0]
self.assertTrue(torch.allclose(post_merge_olayer, pre_merge_olayer))
# test peft model adapter unmerge
model.unmerge_adapter()
post_unmerge_olayer = model(input_tokens)[0]
self.assertTrue(torch.allclose(post_unmerge_olayer, pre_merge_olayer))
# test LoRA merge and unload
model = model.merge_and_unload()
post_unload_merge_olayer = model(input_tokens)[0]
self.assertTrue(torch.allclose(post_unload_merge_olayer, pre_merge_olayer))
@require_torch_gpu
class LoftQTests(unittest.TestCase):
r"""
Tests for LoftQ to ensure that it reduces the quantization error compared to normal LoRA quantization.
"""
def setUp(self):
self.error_factor = 3
def get_input(self, model_id, device):
tokenizer = AutoTokenizer.from_pretrained(model_id)
inputs = tokenizer("All I want is", padding=True, return_tensors="pt")
if device == "cuda":
inputs = inputs.to("cuda")
return inputs
def get_base_model(self, model_id, device, **kwargs):
cls = AutoModelForSeq2SeqLM if "t5" in model_id else AutoModelForCausalLM
model = cls.from_pretrained(model_id, **kwargs).eval()
if device == "cuda":
model = model.to("cuda")
return model
def get_logits(self, model, inputs):
if model.config.is_encoder_decoder:
input_ids = inputs["input_ids"]
return model(input_ids=input_ids, decoder_input_ids=input_ids).logits
return model(**inputs).logits
def get_errors(
self, bits=4, loftq_iter=1, device="cuda", model_id="hf-internal-testing/tiny-random-BloomForCausalLM"
):
# Helper function that returns the quantization errors (MAE and MSE) when comparing the quantized LoRA model
# to the base model, vs the LoftQ quantized model to the base model. We expect the LoftQ quantized model to
# have less error than the normal LoRA quantized model. Since we compare logits, the observed error is
# already somewhat dampened because of the softmax.
torch.manual_seed(0)
model = self.get_base_model(model_id, device)
task_type = TaskType.SEQ_2_SEQ_LM if model.config.is_encoder_decoder else TaskType.CAUSAL_LM
inputs = self.get_input(model_id, device)
logits_base = self.get_logits(model, inputs)
# clean up
del model
gc.collect()
torch.cuda.empty_cache()
# logits from the normal quantized LoRA model
lora_config = LoraConfig(task_type=task_type)
kwargs = {}
if bits == 4:
kwargs["load_in_4bit"] = True
elif bits == 8:
kwargs["load_in_8bit"] = True
else:
raise ValueError("bits must be 4 or 8")
quantized_model = get_peft_model(
self.get_base_model(model_id, device=None, **kwargs),
lora_config,
)
torch.manual_seed(0)
logits_quantized = self.get_logits(quantized_model, inputs)
del quantized_model
gc.collect()
torch.cuda.empty_cache()
# logits from quantized LoRA model using LoftQ
loftq_config = LoftQConfig(loftq_bits=bits, loftq_iter=loftq_iter)
lora_config = LoraConfig(task_type=TaskType.CAUSAL_LM, init_lora_weights="loftq", loftq_config=loftq_config)
model = self.get_base_model(model_id, device)
if device == "cuda":
model = model.to("cuda")
loftq_model = get_peft_model(model, lora_config)
if device == "cuda":
loftq_model = loftq_model.to("cuda")
torch.manual_seed(0)
logits_loftq = self.get_logits(loftq_model, inputs)
del loftq_model
gc.collect()
torch.cuda.empty_cache()
mae_quantized = torch.abs(logits_base - logits_quantized).mean()
mse_quantized = torch.pow(logits_base - logits_quantized, 2).mean()
mae_loftq = torch.abs(logits_base - logits_loftq).mean()
mse_loftq = torch.pow(logits_base - logits_loftq, 2).mean()
return mae_quantized, mse_quantized, mae_loftq, mse_loftq
@parameterized.expand(["cuda", "cpu"])
def test_bloomz_loftq_4bit(self, device):
# In this test, we compare the logits of the base model, the quantized LoRA model, and the quantized model
# using LoftQ. When quantizing, we expect a certain level of error. However, we expect the LoftQ quantized
# model to have less error than the normal LoRA quantized model. Note that when using normal LoRA, the
# quantization error is simply the error from quantization without LoRA, as LoRA is a no-op before training.
# We still apply LoRA for the test for consistency.
mae_quantized, mse_quantized, mae_loftq, mse_loftq = self.get_errors(bits=4, device=device)
# first, sanity check that all errors are > 0.0
self.assertTrue(mae_quantized > 0.0)
self.assertTrue(mse_quantized > 0.0)
self.assertTrue(mae_loftq > 0.0)
self.assertTrue(mse_loftq > 0.0)
# next, check that LoftQ quantization errors are smaller than LoRA errors by a certain margin
factor = 3
self.assertTrue(mae_loftq < mae_quantized / factor)
self.assertTrue(mse_loftq < mse_quantized / factor)
@parameterized.expand(["cuda", "cpu"])
def test_bloomz_loftq_4bit_iter_5(self, device):
# Same test as the previous one but with 5 iterations. We should expect the error to be even smaller with more
# iterations, but in practice the difference is not that large, at least not for this small base model.
mae_quantized, mse_quantized, mae_loftq, mse_loftq = self.get_errors(bits=4, loftq_iter=5, device=device)
# first, sanity check that all errors are > 0.0
self.assertTrue(mae_quantized > 0.0)
self.assertTrue(mse_quantized > 0.0)
self.assertTrue(mae_loftq > 0.0)
self.assertTrue(mse_loftq > 0.0)
# next, check that LoftQ quantization errors are smaller than LoRA errors by a certain margin
self.assertTrue(mae_loftq < mae_quantized / self.error_factor)
self.assertTrue(mse_loftq < mse_quantized / self.error_factor)
@parameterized.expand(["cuda", "cpu"])
def test_bloomz_loftq_8bit(self, device):
# Same test as test_bloomz_loftq_4bit but with 8 bits.
mae_quantized, mse_quantized, mae_loftq, mse_loftq = self.get_errors(bits=8, device=device)
# first, sanity check that all errors are > 0.0
self.assertTrue(mae_quantized > 0.0)
self.assertTrue(mse_quantized > 0.0)
self.assertTrue(mae_loftq > 0.0)
self.assertTrue(mse_loftq > 0.0)
# next, check that LoftQ quantization errors are smaller than LoRA errors by a certain margin
self.assertTrue(mae_loftq < mae_quantized / self.error_factor)
self.assertTrue(mse_loftq < mse_quantized / self.error_factor)
@parameterized.expand(["cuda", "cpu"])
def test_bloomz_loftq_8bit_iter_5(self, device):
# Same test as test_bloomz_loftq_4bit_iter_5 but with 8 bits.
mae_quantized, mse_quantized, mae_loftq, mse_loftq = self.get_errors(bits=8, loftq_iter=5, device=device)
# first, sanity check that all errors are > 0.0
self.assertTrue(mae_quantized > 0.0)
self.assertTrue(mse_quantized > 0.0)
self.assertTrue(mae_loftq > 0.0)
self.assertTrue(mse_loftq > 0.0)
# next, check that LoftQ quantization errors are smaller than LoRA errors by a certain margin
self.assertTrue(mae_loftq < mae_quantized / self.error_factor)
self.assertTrue(mse_loftq < mse_quantized / self.error_factor)
@parameterized.expand(["cuda", "cpu"])
def test_t5_loftq_4bit(self, device):
mae_quantized, mse_quantized, mae_loftq, mse_loftq = self.get_errors(
bits=4, device=device, model_id="t5-small"
)
# first, sanity check that all errors are > 0.0
self.assertTrue(mae_quantized > 0.0)
self.assertTrue(mse_quantized > 0.0)
self.assertTrue(mae_loftq > 0.0)
self.assertTrue(mse_loftq > 0.0)
# next, check that LoftQ quantization errors are smaller than LoRA errors by a certain margin
factor = 3
self.assertTrue(mae_loftq < mae_quantized / factor)
self.assertTrue(mse_loftq < mse_quantized / factor)
@parameterized.expand(["cuda", "cpu"])
def test_t5_loftq_8bit(self, device):
mae_quantized, mse_quantized, mae_loftq, mse_loftq = self.get_errors(
bits=8, device=device, model_id="t5-small"
)
# first, sanity check that all errors are > 0.0
self.assertTrue(mae_quantized > 0.0)
self.assertTrue(mse_quantized > 0.0)
self.assertTrue(mae_loftq > 0.0)
self.assertTrue(mse_loftq > 0.0)
# next, check that LoftQ quantization errors are smaller than LoRA errors by a certain margin
factor = 3
self.assertTrue(mae_loftq < mae_quantized / factor)
self.assertTrue(mse_loftq < mse_quantized / factor)
@require_bitsandbytes
@require_torch_gpu
class MultiprocessTester(unittest.TestCase):
def test_notebook_launcher(self):
script_path = os.path.join("scripts", "launch_notebook_mp.py")
cmd = ["python", script_path]
with patch_environment(omp_num_threads=1):
run_command(cmd, env=os.environ.copy())
@require_torch_gpu
class MixedPrecisionTests(unittest.TestCase):
def setUp(self):
self.causal_lm_model_id = "facebook/opt-350m"
self.tokenizer = AutoTokenizer.from_pretrained(self.causal_lm_model_id)
self.config = LoraConfig(
r=16,
lora_alpha=32,
task_type="CAUSAL_LM",
)
data = load_dataset("ybelkada/english_quotes_copy")
self.data = data.map(lambda samples: self.tokenizer(samples["quote"]), batched=True)
def tearDown(self):
r"""
Efficient mechanism to free GPU memory after each test. Based on
https://github.com/huggingface/transformers/issues/21094
"""
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
gc.collect()
@pytest.mark.single_gpu_tests
def test_model_loaded_in_float16_raises(self):
# This test shows the issue with loading the model in fp16 and then trying to use it with mixed precision
# training, which should not use fp16. If this is ever automated in PEFT, this test should fail. In that case,
# remove this test, adjust the next one, and remove the entry about FP16 usage from troubleshooting.md.
model = AutoModelForCausalLM.from_pretrained(
self.causal_lm_model_id,
torch_dtype=torch.float16,
)
model = get_peft_model(model, self.config)
with tempfile.TemporaryDirectory() as tmp_dir:
trainer = Trainer(
model=model,
train_dataset=self.data["train"],
args=TrainingArguments(
fp16=True, # <= this is required for the error to be raised
logging_steps=1,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(self.tokenizer, mlm=False),
)
msg = "Attempting to unscale FP16 gradients."
with self.assertRaisesRegex(ValueError, msg):
trainer.train()
@pytest.mark.single_gpu_tests
def test_model_loaded_in_float16_working(self):
# Same test as before but containing the fix to make it work
model = AutoModelForCausalLM.from_pretrained(
self.causal_lm_model_id,
torch_dtype=torch.float16,
)
model = get_peft_model(model, self.config)
# for now, this is unfortunately necessary to avoid the error:
# ValueError: Attempting to unscale FP16 gradients.
for param in model.parameters():
if param.requires_grad:
param.data = param.data.float()
with tempfile.TemporaryDirectory() as tmp_dir:
trainer = Trainer(
model=model,
train_dataset=self.data["train"],
args=TrainingArguments(
fp16=True,
max_steps=3,
output_dir=tmp_dir,
),
data_collator=DataCollatorForLanguageModeling(self.tokenizer, mlm=False),
)
trainer.train()
| peft/tests/test_gpu_examples.py/0 | {
"file_path": "peft/tests/test_gpu_examples.py",
"repo_id": "peft",
"token_count": 26064
} | 166 |
title: Model Pages | pytorch-image-models/docs/models/.pages/0 | {
"file_path": "pytorch-image-models/docs/models/.pages",
"repo_id": "pytorch-image-models",
"token_count": 4
} | 167 |
# ESE-VoVNet
**VoVNet** is a convolutional neural network that seeks to make [DenseNet](https://paperswithcode.com/method/densenet) more efficient by concatenating all features only once in the last feature map, which makes input size constant and enables enlarging new output channel.
Read about [one-shot aggregation here](https://paperswithcode.com/method/one-shot-aggregation).
{% include 'code_snippets.md' %}
## How do I train this model?
You can follow the [timm recipe scripts](https://rwightman.github.io/pytorch-image-models/scripts/) for training a new model afresh.
## Citation
```BibTeX
@misc{lee2019energy,
title={An Energy and GPU-Computation Efficient Backbone Network for Real-Time Object Detection},
author={Youngwan Lee and Joong-won Hwang and Sangrok Lee and Yuseok Bae and Jongyoul Park},
year={2019},
eprint={1904.09730},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
<!--
Type: model-index
Collections:
- Name: ESE VovNet
Paper:
Title: 'CenterMask : Real-Time Anchor-Free Instance Segmentation'
URL: https://paperswithcode.com/paper/centermask-real-time-anchor-free-instance-1
Models:
- Name: ese_vovnet19b_dw
In Collection: ESE VovNet
Metadata:
FLOPs: 1711959904
Parameters: 6540000
File Size: 26243175
Architecture:
- Batch Normalization
- Convolution
- Max Pooling
- One-Shot Aggregation
- ReLU
Tasks:
- Image Classification
Training Data:
- ImageNet
ID: ese_vovnet19b_dw
Layers: 19
Crop Pct: '0.875'
Image Size: '224'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/vovnet.py#L361
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/ese_vovnet19b_dw-a8741004.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 76.82%
Top 5 Accuracy: 93.28%
- Name: ese_vovnet39b
In Collection: ESE VovNet
Metadata:
FLOPs: 9089259008
Parameters: 24570000
File Size: 98397138
Architecture:
- Batch Normalization
- Convolution
- Max Pooling
- One-Shot Aggregation
- ReLU
Tasks:
- Image Classification
Training Data:
- ImageNet
ID: ese_vovnet39b
Layers: 39
Crop Pct: '0.875'
Image Size: '224'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/vovnet.py#L371
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/ese_vovnet39b-f912fe73.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 79.31%
Top 5 Accuracy: 94.72%
-->
| pytorch-image-models/docs/models/.templates/models/ese-vovnet.md/0 | {
"file_path": "pytorch-image-models/docs/models/.templates/models/ese-vovnet.md",
"repo_id": "pytorch-image-models",
"token_count": 1127
} | 168 |
# MixNet
**MixNet** is a type of convolutional neural network discovered via AutoML that utilises [MixConvs](https://paperswithcode.com/method/mixconv) instead of regular [depthwise convolutions](https://paperswithcode.com/method/depthwise-convolution).
{% include 'code_snippets.md' %}
## How do I train this model?
You can follow the [timm recipe scripts](https://rwightman.github.io/pytorch-image-models/scripts/) for training a new model afresh.
## Citation
```BibTeX
@misc{tan2019mixconv,
title={MixConv: Mixed Depthwise Convolutional Kernels},
author={Mingxing Tan and Quoc V. Le},
year={2019},
eprint={1907.09595},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
<!--
Type: model-index
Collections:
- Name: MixNet
Paper:
Title: 'MixConv: Mixed Depthwise Convolutional Kernels'
URL: https://paperswithcode.com/paper/mixnet-mixed-depthwise-convolutional-kernels
Models:
- Name: mixnet_l
In Collection: MixNet
Metadata:
FLOPs: 738671316
Parameters: 7330000
File Size: 29608232
Architecture:
- Batch Normalization
- Dense Connections
- Dropout
- Global Average Pooling
- Grouped Convolution
- MixConv
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- MNAS
Training Data:
- ImageNet
ID: mixnet_l
Crop Pct: '0.875'
Image Size: '224'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1669
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mixnet_l-5a9a2ed8.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 78.98%
Top 5 Accuracy: 94.18%
- Name: mixnet_m
In Collection: MixNet
Metadata:
FLOPs: 454543374
Parameters: 5010000
File Size: 20298347
Architecture:
- Batch Normalization
- Dense Connections
- Dropout
- Global Average Pooling
- Grouped Convolution
- MixConv
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- MNAS
Training Data:
- ImageNet
ID: mixnet_m
Crop Pct: '0.875'
Image Size: '224'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1660
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mixnet_m-4647fc68.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 77.27%
Top 5 Accuracy: 93.42%
- Name: mixnet_s
In Collection: MixNet
Metadata:
FLOPs: 321264910
Parameters: 4130000
File Size: 16727982
Architecture:
- Batch Normalization
- Dense Connections
- Dropout
- Global Average Pooling
- Grouped Convolution
- MixConv
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- MNAS
Training Data:
- ImageNet
ID: mixnet_s
Crop Pct: '0.875'
Image Size: '224'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1651
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mixnet_s-a907afbc.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 75.99%
Top 5 Accuracy: 92.79%
- Name: mixnet_xl
In Collection: MixNet
Metadata:
FLOPs: 1195880424
Parameters: 11900000
File Size: 48001170
Architecture:
- Batch Normalization
- Dense Connections
- Dropout
- Global Average Pooling
- Grouped Convolution
- MixConv
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- MNAS
Training Data:
- ImageNet
ID: mixnet_xl
Crop Pct: '0.875'
Image Size: '224'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1678
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mixnet_xl_ra-aac3c00c.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 80.47%
Top 5 Accuracy: 94.93%
-->
| pytorch-image-models/docs/models/.templates/models/mixnet.md/0 | {
"file_path": "pytorch-image-models/docs/models/.templates/models/mixnet.md",
"repo_id": "pytorch-image-models",
"token_count": 1878
} | 169 |
# SE-ResNet
**SE ResNet** is a variant of a [ResNet](https://www.paperswithcode.com/method/resnet) that employs [squeeze-and-excitation blocks](https://paperswithcode.com/method/squeeze-and-excitation-block) to enable the network to perform dynamic channel-wise feature recalibration.
{% include 'code_snippets.md' %}
## How do I train this model?
You can follow the [timm recipe scripts](https://rwightman.github.io/pytorch-image-models/scripts/) for training a new model afresh.
## Citation
```BibTeX
@misc{hu2019squeezeandexcitation,
title={Squeeze-and-Excitation Networks},
author={Jie Hu and Li Shen and Samuel Albanie and Gang Sun and Enhua Wu},
year={2019},
eprint={1709.01507},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
<!--
Type: model-index
Collections:
- Name: SE ResNet
Paper:
Title: Squeeze-and-Excitation Networks
URL: https://paperswithcode.com/paper/squeeze-and-excitation-networks
Models:
- Name: seresnet152d
In Collection: SE ResNet
Metadata:
FLOPs: 20161904304
Parameters: 66840000
File Size: 268144497
Architecture:
- 1x1 Convolution
- Batch Normalization
- Bottleneck Residual Block
- Convolution
- Global Average Pooling
- Max Pooling
- ReLU
- Residual Block
- Residual Connection
- Softmax
- Squeeze-and-Excitation Block
Tasks:
- Image Classification
Training Techniques:
- Label Smoothing
- SGD with Momentum
- Weight Decay
Training Data:
- ImageNet
Training Resources: 8x NVIDIA Titan X GPUs
ID: seresnet152d
LR: 0.6
Epochs: 100
Layers: 152
Dropout: 0.2
Crop Pct: '0.94'
Momentum: 0.9
Batch Size: 1024
Image Size: '256'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/a7f95818e44b281137503bcf4b3e3e94d8ffa52f/timm/models/resnet.py#L1206
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/seresnet152d_ra2-04464dd2.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 83.74%
Top 5 Accuracy: 96.77%
- Name: seresnet50
In Collection: SE ResNet
Metadata:
FLOPs: 5285062320
Parameters: 28090000
File Size: 112621903
Architecture:
- 1x1 Convolution
- Batch Normalization
- Bottleneck Residual Block
- Convolution
- Global Average Pooling
- Max Pooling
- ReLU
- Residual Block
- Residual Connection
- Softmax
- Squeeze-and-Excitation Block
Tasks:
- Image Classification
Training Techniques:
- Label Smoothing
- SGD with Momentum
- Weight Decay
Training Data:
- ImageNet
Training Resources: 8x NVIDIA Titan X GPUs
ID: seresnet50
LR: 0.6
Epochs: 100
Layers: 50
Dropout: 0.2
Crop Pct: '0.875'
Momentum: 0.9
Batch Size: 1024
Image Size: '224'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/a7f95818e44b281137503bcf4b3e3e94d8ffa52f/timm/models/resnet.py#L1180
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/seresnet50_ra_224-8efdb4bb.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 80.26%
Top 5 Accuracy: 95.07%
-->
| pytorch-image-models/docs/models/.templates/models/se-resnet.md/0 | {
"file_path": "pytorch-image-models/docs/models/.templates/models/se-resnet.md",
"repo_id": "pytorch-image-models",
"token_count": 1371
} | 170 |
# TResNet
A **TResNet** is a variant on a [ResNet](https://paperswithcode.com/method/resnet) that aim to boost accuracy while maintaining GPU training and inference efficiency. They contain several design tricks including a SpaceToDepth stem, [Anti-Alias downsampling](https://paperswithcode.com/method/anti-alias-downsampling), In-Place Activated BatchNorm, Blocks selection and [squeeze-and-excitation layers](https://paperswithcode.com/method/squeeze-and-excitation-block).
{% include 'code_snippets.md' %}
## How do I train this model?
You can follow the [timm recipe scripts](https://rwightman.github.io/pytorch-image-models/scripts/) for training a new model afresh.
## Citation
```BibTeX
@misc{ridnik2020tresnet,
title={TResNet: High Performance GPU-Dedicated Architecture},
author={Tal Ridnik and Hussam Lawen and Asaf Noy and Emanuel Ben Baruch and Gilad Sharir and Itamar Friedman},
year={2020},
eprint={2003.13630},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
<!--
Type: model-index
Collections:
- Name: TResNet
Paper:
Title: 'TResNet: High Performance GPU-Dedicated Architecture'
URL: https://paperswithcode.com/paper/tresnet-high-performance-gpu-dedicated
Models:
- Name: tresnet_l
In Collection: TResNet
Metadata:
FLOPs: 10873416792
Parameters: 53456696
File Size: 224440219
Architecture:
- 1x1 Convolution
- Anti-Alias Downsampling
- Convolution
- Global Average Pooling
- InPlace-ABN
- Leaky ReLU
- ReLU
- Residual Connection
- Squeeze-and-Excitation Block
Tasks:
- Image Classification
Training Techniques:
- AutoAugment
- Cutout
- Label Smoothing
- SGD with Momentum
- Weight Decay
Training Data:
- ImageNet
Training Resources: 8x NVIDIA 100 GPUs
ID: tresnet_l
LR: 0.01
Epochs: 300
Crop Pct: '0.875'
Momentum: 0.9
Image Size: '224'
Weight Decay: 0.0001
Interpolation: bilinear
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/tresnet.py#L267
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tresnet/tresnet_l_81_5-235b486c.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 81.49%
Top 5 Accuracy: 95.62%
- Name: tresnet_l_448
In Collection: TResNet
Metadata:
FLOPs: 43488238584
Parameters: 53456696
File Size: 224440219
Architecture:
- 1x1 Convolution
- Anti-Alias Downsampling
- Convolution
- Global Average Pooling
- InPlace-ABN
- Leaky ReLU
- ReLU
- Residual Connection
- Squeeze-and-Excitation Block
Tasks:
- Image Classification
Training Techniques:
- AutoAugment
- Cutout
- Label Smoothing
- SGD with Momentum
- Weight Decay
Training Data:
- ImageNet
Training Resources: 8x NVIDIA 100 GPUs
ID: tresnet_l_448
LR: 0.01
Epochs: 300
Crop Pct: '0.875'
Momentum: 0.9
Image Size: '448'
Weight Decay: 0.0001
Interpolation: bilinear
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/tresnet.py#L285
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tresnet/tresnet_l_448-940d0cd1.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 82.26%
Top 5 Accuracy: 95.98%
- Name: tresnet_m
In Collection: TResNet
Metadata:
FLOPs: 5733048064
Parameters: 41282200
File Size: 125861314
Architecture:
- 1x1 Convolution
- Anti-Alias Downsampling
- Convolution
- Global Average Pooling
- InPlace-ABN
- Leaky ReLU
- ReLU
- Residual Connection
- Squeeze-and-Excitation Block
Tasks:
- Image Classification
Training Techniques:
- AutoAugment
- Cutout
- Label Smoothing
- SGD with Momentum
- Weight Decay
Training Data:
- ImageNet
Training Resources: 8x NVIDIA 100 GPUs
Training Time: < 24 hours
ID: tresnet_m
LR: 0.01
Epochs: 300
Crop Pct: '0.875'
Momentum: 0.9
Image Size: '224'
Weight Decay: 0.0001
Interpolation: bilinear
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/tresnet.py#L261
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tresnet/tresnet_m_80_8-dbc13962.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 80.8%
Top 5 Accuracy: 94.86%
- Name: tresnet_m_448
In Collection: TResNet
Metadata:
FLOPs: 22929743104
Parameters: 29278464
File Size: 125861314
Architecture:
- 1x1 Convolution
- Anti-Alias Downsampling
- Convolution
- Global Average Pooling
- InPlace-ABN
- Leaky ReLU
- ReLU
- Residual Connection
- Squeeze-and-Excitation Block
Tasks:
- Image Classification
Training Techniques:
- AutoAugment
- Cutout
- Label Smoothing
- SGD with Momentum
- Weight Decay
Training Data:
- ImageNet
Training Resources: 8x NVIDIA 100 GPUs
ID: tresnet_m_448
LR: 0.01
Epochs: 300
Crop Pct: '0.875'
Momentum: 0.9
Image Size: '448'
Weight Decay: 0.0001
Interpolation: bilinear
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/tresnet.py#L279
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tresnet/tresnet_m_448-bc359d10.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 81.72%
Top 5 Accuracy: 95.57%
- Name: tresnet_xl
In Collection: TResNet
Metadata:
FLOPs: 15162534034
Parameters: 75646610
File Size: 314378965
Architecture:
- 1x1 Convolution
- Anti-Alias Downsampling
- Convolution
- Global Average Pooling
- InPlace-ABN
- Leaky ReLU
- ReLU
- Residual Connection
- Squeeze-and-Excitation Block
Tasks:
- Image Classification
Training Techniques:
- AutoAugment
- Cutout
- Label Smoothing
- SGD with Momentum
- Weight Decay
Training Data:
- ImageNet
Training Resources: 8x NVIDIA 100 GPUs
ID: tresnet_xl
LR: 0.01
Epochs: 300
Crop Pct: '0.875'
Momentum: 0.9
Image Size: '224'
Weight Decay: 0.0001
Interpolation: bilinear
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/tresnet.py#L273
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tresnet/tresnet_xl_82_0-a2d51b00.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 82.05%
Top 5 Accuracy: 95.93%
- Name: tresnet_xl_448
In Collection: TResNet
Metadata:
FLOPs: 60641712730
Parameters: 75646610
File Size: 224440219
Architecture:
- 1x1 Convolution
- Anti-Alias Downsampling
- Convolution
- Global Average Pooling
- InPlace-ABN
- Leaky ReLU
- ReLU
- Residual Connection
- Squeeze-and-Excitation Block
Tasks:
- Image Classification
Training Techniques:
- AutoAugment
- Cutout
- Label Smoothing
- SGD with Momentum
- Weight Decay
Training Data:
- ImageNet
Training Resources: 8x NVIDIA 100 GPUs
ID: tresnet_xl_448
LR: 0.01
Epochs: 300
Crop Pct: '0.875'
Momentum: 0.9
Image Size: '448'
Weight Decay: 0.0001
Interpolation: bilinear
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/tresnet.py#L291
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-tresnet/tresnet_l_448-940d0cd1.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 83.06%
Top 5 Accuracy: 96.19%
-->
| pytorch-image-models/docs/models/.templates/models/tresnet.md/0 | {
"file_path": "pytorch-image-models/docs/models/.templates/models/tresnet.md",
"repo_id": "pytorch-image-models",
"token_count": 3391
} | 171 |
# # Ensemble Adversarial Inception ResNet v2
**Inception-ResNet-v2** is a convolutional neural architecture that builds on the Inception family of architectures but incorporates [residual connections](https://paperswithcode.com/method/residual-connection) (replacing the filter concatenation stage of the Inception architecture).
This particular model was trained for study of adversarial examples (adversarial training).
The weights from this model were ported from [Tensorflow/Models](https://github.com/tensorflow/models).
## How do I use this model on an image?
To load a pretrained model:
```python
import timm
model = timm.create_model('ens_adv_inception_resnet_v2', pretrained=True)
model.eval()
```
To load and preprocess the image:
```python
import urllib
from PIL import Image
from timm.data import resolve_data_config
from timm.data.transforms_factory import create_transform
config = resolve_data_config({}, model=model)
transform = create_transform(**config)
url, filename = ("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")
urllib.request.urlretrieve(url, filename)
img = Image.open(filename).convert('RGB')
tensor = transform(img).unsqueeze(0) # transform and add batch dimension
```
To get the model predictions:
```python
import torch
with torch.no_grad():
out = model(tensor)
probabilities = torch.nn.functional.softmax(out[0], dim=0)
print(probabilities.shape)
# prints: torch.Size([1000])
```
To get the top-5 predictions class names:
```python
# Get imagenet class mappings
url, filename = ("https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt", "imagenet_classes.txt")
urllib.request.urlretrieve(url, filename)
with open("imagenet_classes.txt", "r") as f:
categories = [s.strip() for s in f.readlines()]
# Print top categories per image
top5_prob, top5_catid = torch.topk(probabilities, 5)
for i in range(top5_prob.size(0)):
print(categories[top5_catid[i]], top5_prob[i].item())
# prints class names and probabilities like:
# [('Samoyed', 0.6425196528434753), ('Pomeranian', 0.04062102362513542), ('keeshond', 0.03186424449086189), ('white wolf', 0.01739676296710968), ('Eskimo dog', 0.011717947199940681)]
```
Replace the model name with the variant you want to use, e.g. `ens_adv_inception_resnet_v2`. You can find the IDs in the model summaries at the top of this page.
To extract image features with this model, follow the [timm feature extraction examples](https://rwightman.github.io/pytorch-image-models/feature_extraction/), just change the name of the model you want to use.
## How do I finetune this model?
You can finetune any of the pre-trained models just by changing the classifier (the last layer).
```python
model = timm.create_model('ens_adv_inception_resnet_v2', pretrained=True, num_classes=NUM_FINETUNE_CLASSES)
```
To finetune on your own dataset, you have to write a training loop or adapt [timm's training
script](https://github.com/rwightman/pytorch-image-models/blob/master/train.py) to use your dataset.
## How do I train this model?
You can follow the [timm recipe scripts](https://rwightman.github.io/pytorch-image-models/scripts/) for training a new model afresh.
## Citation
```BibTeX
@article{DBLP:journals/corr/abs-1804-00097,
author = {Alexey Kurakin and
Ian J. Goodfellow and
Samy Bengio and
Yinpeng Dong and
Fangzhou Liao and
Ming Liang and
Tianyu Pang and
Jun Zhu and
Xiaolin Hu and
Cihang Xie and
Jianyu Wang and
Zhishuai Zhang and
Zhou Ren and
Alan L. Yuille and
Sangxia Huang and
Yao Zhao and
Yuzhe Zhao and
Zhonglin Han and
Junjiajia Long and
Yerkebulan Berdibekov and
Takuya Akiba and
Seiya Tokui and
Motoki Abe},
title = {Adversarial Attacks and Defences Competition},
journal = {CoRR},
volume = {abs/1804.00097},
year = {2018},
url = {http://arxiv.org/abs/1804.00097},
archivePrefix = {arXiv},
eprint = {1804.00097},
timestamp = {Thu, 31 Oct 2019 16:31:22 +0100},
biburl = {https://dblp.org/rec/journals/corr/abs-1804-00097.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<!--
Type: model-index
Collections:
- Name: Ensemble Adversarial
Paper:
Title: Adversarial Attacks and Defences Competition
URL: https://paperswithcode.com/paper/adversarial-attacks-and-defences-competition
Models:
- Name: ens_adv_inception_resnet_v2
In Collection: Ensemble Adversarial
Metadata:
FLOPs: 16959133120
Parameters: 55850000
File Size: 223774238
Architecture:
- 1x1 Convolution
- Auxiliary Classifier
- Average Pooling
- Average Pooling
- Batch Normalization
- Convolution
- Dense Connections
- Dropout
- Inception-v3 Module
- Max Pooling
- ReLU
- Softmax
Tasks:
- Image Classification
Training Data:
- ImageNet
ID: ens_adv_inception_resnet_v2
Crop Pct: '0.897'
Image Size: '299'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/inception_resnet_v2.py#L351
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/ens_adv_inception_resnet_v2-2592a550.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 1.0%
Top 5 Accuracy: 17.32%
--> | pytorch-image-models/docs/models/ensemble-adversarial.md/0 | {
"file_path": "pytorch-image-models/docs/models/ensemble-adversarial.md",
"repo_id": "pytorch-image-models",
"token_count": 2206
} | 172 |
# RexNet
**Rank Expansion Networks** (ReXNets) follow a set of new design principles for designing bottlenecks in image classification models. Authors refine each layer by 1) expanding the input channel size of the convolution layer and 2) replacing the [ReLU6s](https://www.paperswithcode.com/method/relu6).
## How do I use this model on an image?
To load a pretrained model:
```python
import timm
model = timm.create_model('rexnet_100', pretrained=True)
model.eval()
```
To load and preprocess the image:
```python
import urllib
from PIL import Image
from timm.data import resolve_data_config
from timm.data.transforms_factory import create_transform
config = resolve_data_config({}, model=model)
transform = create_transform(**config)
url, filename = ("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")
urllib.request.urlretrieve(url, filename)
img = Image.open(filename).convert('RGB')
tensor = transform(img).unsqueeze(0) # transform and add batch dimension
```
To get the model predictions:
```python
import torch
with torch.no_grad():
out = model(tensor)
probabilities = torch.nn.functional.softmax(out[0], dim=0)
print(probabilities.shape)
# prints: torch.Size([1000])
```
To get the top-5 predictions class names:
```python
# Get imagenet class mappings
url, filename = ("https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt", "imagenet_classes.txt")
urllib.request.urlretrieve(url, filename)
with open("imagenet_classes.txt", "r") as f:
categories = [s.strip() for s in f.readlines()]
# Print top categories per image
top5_prob, top5_catid = torch.topk(probabilities, 5)
for i in range(top5_prob.size(0)):
print(categories[top5_catid[i]], top5_prob[i].item())
# prints class names and probabilities like:
# [('Samoyed', 0.6425196528434753), ('Pomeranian', 0.04062102362513542), ('keeshond', 0.03186424449086189), ('white wolf', 0.01739676296710968), ('Eskimo dog', 0.011717947199940681)]
```
Replace the model name with the variant you want to use, e.g. `rexnet_100`. You can find the IDs in the model summaries at the top of this page.
To extract image features with this model, follow the [timm feature extraction examples](https://rwightman.github.io/pytorch-image-models/feature_extraction/), just change the name of the model you want to use.
## How do I finetune this model?
You can finetune any of the pre-trained models just by changing the classifier (the last layer).
```python
model = timm.create_model('rexnet_100', pretrained=True, num_classes=NUM_FINETUNE_CLASSES)
```
To finetune on your own dataset, you have to write a training loop or adapt [timm's training
script](https://github.com/rwightman/pytorch-image-models/blob/master/train.py) to use your dataset.
## How do I train this model?
You can follow the [timm recipe scripts](https://rwightman.github.io/pytorch-image-models/scripts/) for training a new model afresh.
## Citation
```BibTeX
@misc{han2020rexnet,
title={ReXNet: Diminishing Representational Bottleneck on Convolutional Neural Network},
author={Dongyoon Han and Sangdoo Yun and Byeongho Heo and YoungJoon Yoo},
year={2020},
eprint={2007.00992},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
<!--
Type: model-index
Collections:
- Name: RexNet
Paper:
Title: 'ReXNet: Diminishing Representational Bottleneck on Convolutional Neural
Network'
URL: https://paperswithcode.com/paper/rexnet-diminishing-representational
Models:
- Name: rexnet_100
In Collection: RexNet
Metadata:
FLOPs: 509989377
Parameters: 4800000
File Size: 19417552
Architecture:
- Batch Normalization
- Convolution
- Dropout
- ReLU6
- Residual Connection
Tasks:
- Image Classification
Training Techniques:
- Label Smoothing
- Linear Warmup With Cosine Annealing
- Nesterov Accelerated Gradient
- Weight Decay
Training Data:
- ImageNet
Training Resources: 4x NVIDIA V100 GPUs
ID: rexnet_100
LR: 0.5
Epochs: 400
Dropout: 0.2
Crop Pct: '0.875'
Momentum: 0.9
Batch Size: 512
Image Size: '224'
Weight Decay: 1.0e-05
Interpolation: bicubic
Label Smoothing: 0.1
Code: https://github.com/rwightman/pytorch-image-models/blob/b9843f954b0457af2db4f9dea41a8538f51f5d78/timm/models/rexnet.py#L212
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rexnet/rexnetv1_100-1b4dddf4.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 77.86%
Top 5 Accuracy: 93.88%
- Name: rexnet_130
In Collection: RexNet
Metadata:
FLOPs: 848364461
Parameters: 7560000
File Size: 30508197
Architecture:
- Batch Normalization
- Convolution
- Dropout
- ReLU6
- Residual Connection
Tasks:
- Image Classification
Training Techniques:
- Label Smoothing
- Linear Warmup With Cosine Annealing
- Nesterov Accelerated Gradient
- Weight Decay
Training Data:
- ImageNet
Training Resources: 4x NVIDIA V100 GPUs
ID: rexnet_130
LR: 0.5
Epochs: 400
Dropout: 0.2
Crop Pct: '0.875'
Momentum: 0.9
Batch Size: 512
Image Size: '224'
Weight Decay: 1.0e-05
Interpolation: bicubic
Label Smoothing: 0.1
Code: https://github.com/rwightman/pytorch-image-models/blob/b9843f954b0457af2db4f9dea41a8538f51f5d78/timm/models/rexnet.py#L218
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rexnet/rexnetv1_130-590d768e.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 79.49%
Top 5 Accuracy: 94.67%
- Name: rexnet_150
In Collection: RexNet
Metadata:
FLOPs: 1122374469
Parameters: 9730000
File Size: 39227315
Architecture:
- Batch Normalization
- Convolution
- Dropout
- ReLU6
- Residual Connection
Tasks:
- Image Classification
Training Techniques:
- Label Smoothing
- Linear Warmup With Cosine Annealing
- Nesterov Accelerated Gradient
- Weight Decay
Training Data:
- ImageNet
Training Resources: 4x NVIDIA V100 GPUs
ID: rexnet_150
LR: 0.5
Epochs: 400
Dropout: 0.2
Crop Pct: '0.875'
Momentum: 0.9
Batch Size: 512
Image Size: '224'
Weight Decay: 1.0e-05
Interpolation: bicubic
Label Smoothing: 0.1
Code: https://github.com/rwightman/pytorch-image-models/blob/b9843f954b0457af2db4f9dea41a8538f51f5d78/timm/models/rexnet.py#L224
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rexnet/rexnetv1_150-bd1a6aa8.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 80.31%
Top 5 Accuracy: 95.16%
- Name: rexnet_200
In Collection: RexNet
Metadata:
FLOPs: 1960224938
Parameters: 16370000
File Size: 65862221
Architecture:
- Batch Normalization
- Convolution
- Dropout
- ReLU6
- Residual Connection
Tasks:
- Image Classification
Training Techniques:
- Label Smoothing
- Linear Warmup With Cosine Annealing
- Nesterov Accelerated Gradient
- Weight Decay
Training Data:
- ImageNet
Training Resources: 4x NVIDIA V100 GPUs
ID: rexnet_200
LR: 0.5
Epochs: 400
Dropout: 0.2
Crop Pct: '0.875'
Momentum: 0.9
Batch Size: 512
Image Size: '224'
Weight Decay: 1.0e-05
Interpolation: bicubic
Label Smoothing: 0.1
Code: https://github.com/rwightman/pytorch-image-models/blob/b9843f954b0457af2db4f9dea41a8538f51f5d78/timm/models/rexnet.py#L230
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rexnet/rexnetv1_200-8c0b7f2d.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 81.63%
Top 5 Accuracy: 95.67%
--> | pytorch-image-models/docs/models/rexnet.md/0 | {
"file_path": "pytorch-image-models/docs/models/rexnet.md",
"repo_id": "pytorch-image-models",
"token_count": 3081
} | 173 |
# AdvProp (EfficientNet)
**AdvProp** is an adversarial training scheme which treats adversarial examples as additional examples, to prevent overfitting. Key to the method is the usage of a separate auxiliary batch norm for adversarial examples, as they have different underlying distributions to normal examples.
The weights from this model were ported from [Tensorflow/TPU](https://github.com/tensorflow/tpu).
## How do I use this model on an image?
To load a pretrained model:
```py
>>> import timm
>>> model = timm.create_model('tf_efficientnet_b0_ap', pretrained=True)
>>> model.eval()
```
To load and preprocess the image:
```py
>>> import urllib
>>> from PIL import Image
>>> from timm.data import resolve_data_config
>>> from timm.data.transforms_factory import create_transform
>>> config = resolve_data_config({}, model=model)
>>> transform = create_transform(**config)
>>> url, filename = ("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")
>>> urllib.request.urlretrieve(url, filename)
>>> img = Image.open(filename).convert('RGB')
>>> tensor = transform(img).unsqueeze(0) # transform and add batch dimension
```
To get the model predictions:
```py
>>> import torch
>>> with torch.no_grad():
... out = model(tensor)
>>> probabilities = torch.nn.functional.softmax(out[0], dim=0)
>>> print(probabilities.shape)
>>> # prints: torch.Size([1000])
```
To get the top-5 predictions class names:
```py
>>> # Get imagenet class mappings
>>> url, filename = ("https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt", "imagenet_classes.txt")
>>> urllib.request.urlretrieve(url, filename)
>>> with open("imagenet_classes.txt", "r") as f:
... categories = [s.strip() for s in f.readlines()]
>>> # Print top categories per image
>>> top5_prob, top5_catid = torch.topk(probabilities, 5)
>>> for i in range(top5_prob.size(0)):
... print(categories[top5_catid[i]], top5_prob[i].item())
>>> # prints class names and probabilities like:
>>> # [('Samoyed', 0.6425196528434753), ('Pomeranian', 0.04062102362513542), ('keeshond', 0.03186424449086189), ('white wolf', 0.01739676296710968), ('Eskimo dog', 0.011717947199940681)]
```
Replace the model name with the variant you want to use, e.g. `tf_efficientnet_b0_ap`. You can find the IDs in the model summaries at the top of this page.
To extract image features with this model, follow the [timm feature extraction examples](../feature_extraction), just change the name of the model you want to use.
## How do I finetune this model?
You can finetune any of the pre-trained models just by changing the classifier (the last layer).
```py
>>> model = timm.create_model('tf_efficientnet_b0_ap', pretrained=True, num_classes=NUM_FINETUNE_CLASSES)
```
To finetune on your own dataset, you have to write a training loop or adapt [timm's training
script](https://github.com/rwightman/pytorch-image-models/blob/master/train.py) to use your dataset.
## How do I train this model?
You can follow the [timm recipe scripts](../scripts) for training a new model afresh.
## Citation
```BibTeX
@misc{xie2020adversarial,
title={Adversarial Examples Improve Image Recognition},
author={Cihang Xie and Mingxing Tan and Boqing Gong and Jiang Wang and Alan Yuille and Quoc V. Le},
year={2020},
eprint={1911.09665},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
<!--
Type: model-index
Collections:
- Name: AdvProp
Paper:
Title: Adversarial Examples Improve Image Recognition
URL: https://paperswithcode.com/paper/adversarial-examples-improve-image
Models:
- Name: tf_efficientnet_b0_ap
In Collection: AdvProp
Metadata:
FLOPs: 488688572
Parameters: 5290000
File Size: 21385973
Architecture:
- 1x1 Convolution
- Average Pooling
- Batch Normalization
- Convolution
- Dense Connections
- Dropout
- Inverted Residual Block
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- AdvProp
- AutoAugment
- Label Smoothing
- RMSProp
- Stochastic Depth
- Weight Decay
Training Data:
- ImageNet
ID: tf_efficientnet_b0_ap
LR: 0.256
Epochs: 350
Crop Pct: '0.875'
Momentum: 0.9
Batch Size: 2048
Image Size: '224'
Weight Decay: 1.0e-05
Interpolation: bicubic
RMSProp Decay: 0.9
Label Smoothing: 0.1
BatchNorm Momentum: 0.99
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1334
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b0_ap-f262efe1.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 77.1%
Top 5 Accuracy: 93.26%
- Name: tf_efficientnet_b1_ap
In Collection: AdvProp
Metadata:
FLOPs: 883633200
Parameters: 7790000
File Size: 31515350
Architecture:
- 1x1 Convolution
- Average Pooling
- Batch Normalization
- Convolution
- Dense Connections
- Dropout
- Inverted Residual Block
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- AdvProp
- AutoAugment
- Label Smoothing
- RMSProp
- Stochastic Depth
- Weight Decay
Training Data:
- ImageNet
ID: tf_efficientnet_b1_ap
LR: 0.256
Epochs: 350
Crop Pct: '0.882'
Momentum: 0.9
Batch Size: 2048
Image Size: '240'
Weight Decay: 1.0e-05
Interpolation: bicubic
RMSProp Decay: 0.9
Label Smoothing: 0.1
BatchNorm Momentum: 0.99
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1344
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b1_ap-44ef0a3d.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 79.28%
Top 5 Accuracy: 94.3%
- Name: tf_efficientnet_b2_ap
In Collection: AdvProp
Metadata:
FLOPs: 1234321170
Parameters: 9110000
File Size: 36800745
Architecture:
- 1x1 Convolution
- Average Pooling
- Batch Normalization
- Convolution
- Dense Connections
- Dropout
- Inverted Residual Block
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- AdvProp
- AutoAugment
- Label Smoothing
- RMSProp
- Stochastic Depth
- Weight Decay
Training Data:
- ImageNet
ID: tf_efficientnet_b2_ap
LR: 0.256
Epochs: 350
Crop Pct: '0.89'
Momentum: 0.9
Batch Size: 2048
Image Size: '260'
Weight Decay: 1.0e-05
Interpolation: bicubic
RMSProp Decay: 0.9
Label Smoothing: 0.1
BatchNorm Momentum: 0.99
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1354
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b2_ap-2f8e7636.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 80.3%
Top 5 Accuracy: 95.03%
- Name: tf_efficientnet_b3_ap
In Collection: AdvProp
Metadata:
FLOPs: 2275247568
Parameters: 12230000
File Size: 49384538
Architecture:
- 1x1 Convolution
- Average Pooling
- Batch Normalization
- Convolution
- Dense Connections
- Dropout
- Inverted Residual Block
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- AdvProp
- AutoAugment
- Label Smoothing
- RMSProp
- Stochastic Depth
- Weight Decay
Training Data:
- ImageNet
ID: tf_efficientnet_b3_ap
LR: 0.256
Epochs: 350
Crop Pct: '0.904'
Momentum: 0.9
Batch Size: 2048
Image Size: '300'
Weight Decay: 1.0e-05
Interpolation: bicubic
RMSProp Decay: 0.9
Label Smoothing: 0.1
BatchNorm Momentum: 0.99
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1364
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b3_ap-aad25bdd.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 81.82%
Top 5 Accuracy: 95.62%
- Name: tf_efficientnet_b4_ap
In Collection: AdvProp
Metadata:
FLOPs: 5749638672
Parameters: 19340000
File Size: 77993585
Architecture:
- 1x1 Convolution
- Average Pooling
- Batch Normalization
- Convolution
- Dense Connections
- Dropout
- Inverted Residual Block
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- AdvProp
- AutoAugment
- Label Smoothing
- RMSProp
- Stochastic Depth
- Weight Decay
Training Data:
- ImageNet
ID: tf_efficientnet_b4_ap
LR: 0.256
Epochs: 350
Crop Pct: '0.922'
Momentum: 0.9
Batch Size: 2048
Image Size: '380'
Weight Decay: 1.0e-05
Interpolation: bicubic
RMSProp Decay: 0.9
Label Smoothing: 0.1
BatchNorm Momentum: 0.99
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1374
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b4_ap-dedb23e6.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 83.26%
Top 5 Accuracy: 96.39%
- Name: tf_efficientnet_b5_ap
In Collection: AdvProp
Metadata:
FLOPs: 13176501888
Parameters: 30390000
File Size: 122403150
Architecture:
- 1x1 Convolution
- Average Pooling
- Batch Normalization
- Convolution
- Dense Connections
- Dropout
- Inverted Residual Block
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- AdvProp
- AutoAugment
- Label Smoothing
- RMSProp
- Stochastic Depth
- Weight Decay
Training Data:
- ImageNet
ID: tf_efficientnet_b5_ap
LR: 0.256
Epochs: 350
Crop Pct: '0.934'
Momentum: 0.9
Batch Size: 2048
Image Size: '456'
Weight Decay: 1.0e-05
Interpolation: bicubic
RMSProp Decay: 0.9
Label Smoothing: 0.1
BatchNorm Momentum: 0.99
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1384
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b5_ap-9e82fae8.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 84.25%
Top 5 Accuracy: 96.97%
- Name: tf_efficientnet_b6_ap
In Collection: AdvProp
Metadata:
FLOPs: 24180518488
Parameters: 43040000
File Size: 173237466
Architecture:
- 1x1 Convolution
- Average Pooling
- Batch Normalization
- Convolution
- Dense Connections
- Dropout
- Inverted Residual Block
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- AdvProp
- AutoAugment
- Label Smoothing
- RMSProp
- Stochastic Depth
- Weight Decay
Training Data:
- ImageNet
ID: tf_efficientnet_b6_ap
LR: 0.256
Epochs: 350
Crop Pct: '0.942'
Momentum: 0.9
Batch Size: 2048
Image Size: '528'
Weight Decay: 1.0e-05
Interpolation: bicubic
RMSProp Decay: 0.9
Label Smoothing: 0.1
BatchNorm Momentum: 0.99
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1394
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b6_ap-4ffb161f.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 84.79%
Top 5 Accuracy: 97.14%
- Name: tf_efficientnet_b7_ap
In Collection: AdvProp
Metadata:
FLOPs: 48205304880
Parameters: 66349999
File Size: 266850607
Architecture:
- 1x1 Convolution
- Average Pooling
- Batch Normalization
- Convolution
- Dense Connections
- Dropout
- Inverted Residual Block
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- AdvProp
- AutoAugment
- Label Smoothing
- RMSProp
- Stochastic Depth
- Weight Decay
Training Data:
- ImageNet
ID: tf_efficientnet_b7_ap
LR: 0.256
Epochs: 350
Crop Pct: '0.949'
Momentum: 0.9
Batch Size: 2048
Image Size: '600'
Weight Decay: 1.0e-05
Interpolation: bicubic
RMSProp Decay: 0.9
Label Smoothing: 0.1
BatchNorm Momentum: 0.99
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1405
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b7_ap-ddb28fec.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 85.12%
Top 5 Accuracy: 97.25%
- Name: tf_efficientnet_b8_ap
In Collection: AdvProp
Metadata:
FLOPs: 80962956270
Parameters: 87410000
File Size: 351412563
Architecture:
- 1x1 Convolution
- Average Pooling
- Batch Normalization
- Convolution
- Dense Connections
- Dropout
- Inverted Residual Block
- Squeeze-and-Excitation Block
- Swish
Tasks:
- Image Classification
Training Techniques:
- AdvProp
- AutoAugment
- Label Smoothing
- RMSProp
- Stochastic Depth
- Weight Decay
Training Data:
- ImageNet
ID: tf_efficientnet_b8_ap
LR: 0.128
Epochs: 350
Crop Pct: '0.954'
Momentum: 0.9
Batch Size: 2048
Image Size: '672'
Weight Decay: 1.0e-05
Interpolation: bicubic
RMSProp Decay: 0.9
Label Smoothing: 0.1
BatchNorm Momentum: 0.99
Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/efficientnet.py#L1416
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_efficientnet_b8_ap-00e169fa.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 85.37%
Top 5 Accuracy: 97.3%
--> | pytorch-image-models/hfdocs/source/models/advprop.mdx/0 | {
"file_path": "pytorch-image-models/hfdocs/source/models/advprop.mdx",
"repo_id": "pytorch-image-models",
"token_count": 6032
} | 174 |
# (Gluon) ResNeXt
A **ResNeXt** repeats a [building block](https://paperswithcode.com/method/resnext-block) that aggregates a set of transformations with the same topology. Compared to a [ResNet](https://paperswithcode.com/method/resnet), it exposes a new dimension, *cardinality* (the size of the set of transformations) \\( C \\), as an essential factor in addition to the dimensions of depth and width.
The weights from this model were ported from [Gluon](https://cv.gluon.ai/model_zoo/classification.html).
## How do I use this model on an image?
To load a pretrained model:
```py
>>> import timm
>>> model = timm.create_model('gluon_resnext101_32x4d', pretrained=True)
>>> model.eval()
```
To load and preprocess the image:
```py
>>> import urllib
>>> from PIL import Image
>>> from timm.data import resolve_data_config
>>> from timm.data.transforms_factory import create_transform
>>> config = resolve_data_config({}, model=model)
>>> transform = create_transform(**config)
>>> url, filename = ("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")
>>> urllib.request.urlretrieve(url, filename)
>>> img = Image.open(filename).convert('RGB')
>>> tensor = transform(img).unsqueeze(0) # transform and add batch dimension
```
To get the model predictions:
```py
>>> import torch
>>> with torch.no_grad():
... out = model(tensor)
>>> probabilities = torch.nn.functional.softmax(out[0], dim=0)
>>> print(probabilities.shape)
>>> # prints: torch.Size([1000])
```
To get the top-5 predictions class names:
```py
>>> # Get imagenet class mappings
>>> url, filename = ("https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt", "imagenet_classes.txt")
>>> urllib.request.urlretrieve(url, filename)
>>> with open("imagenet_classes.txt", "r") as f:
... categories = [s.strip() for s in f.readlines()]
>>> # Print top categories per image
>>> top5_prob, top5_catid = torch.topk(probabilities, 5)
>>> for i in range(top5_prob.size(0)):
... print(categories[top5_catid[i]], top5_prob[i].item())
>>> # prints class names and probabilities like:
>>> # [('Samoyed', 0.6425196528434753), ('Pomeranian', 0.04062102362513542), ('keeshond', 0.03186424449086189), ('white wolf', 0.01739676296710968), ('Eskimo dog', 0.011717947199940681)]
```
Replace the model name with the variant you want to use, e.g. `gluon_resnext101_32x4d`. You can find the IDs in the model summaries at the top of this page.
To extract image features with this model, follow the [timm feature extraction examples](../feature_extraction), just change the name of the model you want to use.
## How do I finetune this model?
You can finetune any of the pre-trained models just by changing the classifier (the last layer).
```py
>>> model = timm.create_model('gluon_resnext101_32x4d', pretrained=True, num_classes=NUM_FINETUNE_CLASSES)
```
To finetune on your own dataset, you have to write a training loop or adapt [timm's training
script](https://github.com/rwightman/pytorch-image-models/blob/master/train.py) to use your dataset.
## How do I train this model?
You can follow the [timm recipe scripts](../scripts) for training a new model afresh.
## Citation
```BibTeX
@article{DBLP:journals/corr/XieGDTH16,
author = {Saining Xie and
Ross B. Girshick and
Piotr Doll{\'{a}}r and
Zhuowen Tu and
Kaiming He},
title = {Aggregated Residual Transformations for Deep Neural Networks},
journal = {CoRR},
volume = {abs/1611.05431},
year = {2016},
url = {http://arxiv.org/abs/1611.05431},
archivePrefix = {arXiv},
eprint = {1611.05431},
timestamp = {Mon, 13 Aug 2018 16:45:58 +0200},
biburl = {https://dblp.org/rec/journals/corr/XieGDTH16.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<!--
Type: model-index
Collections:
- Name: Gloun ResNeXt
Paper:
Title: Aggregated Residual Transformations for Deep Neural Networks
URL: https://paperswithcode.com/paper/aggregated-residual-transformations-for-deep
Models:
- Name: gluon_resnext101_32x4d
In Collection: Gloun ResNeXt
Metadata:
FLOPs: 10298145792
Parameters: 44180000
File Size: 177367414
Architecture:
- 1x1 Convolution
- Batch Normalization
- Convolution
- Global Average Pooling
- Grouped Convolution
- Max Pooling
- ReLU
- ResNeXt Block
- Residual Connection
- Softmax
Tasks:
- Image Classification
Training Data:
- ImageNet
ID: gluon_resnext101_32x4d
Crop Pct: '0.875'
Image Size: '224'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/gluon_resnet.py#L193
Weights: https://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnext101_32x4d-b253c8c4.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 80.33%
Top 5 Accuracy: 94.91%
- Name: gluon_resnext101_64x4d
In Collection: Gloun ResNeXt
Metadata:
FLOPs: 19954172928
Parameters: 83460000
File Size: 334737852
Architecture:
- 1x1 Convolution
- Batch Normalization
- Convolution
- Global Average Pooling
- Grouped Convolution
- Max Pooling
- ReLU
- ResNeXt Block
- Residual Connection
- Softmax
Tasks:
- Image Classification
Training Data:
- ImageNet
ID: gluon_resnext101_64x4d
Crop Pct: '0.875'
Image Size: '224'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/gluon_resnet.py#L201
Weights: https://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnext101_64x4d-f9a8e184.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 80.63%
Top 5 Accuracy: 95.0%
- Name: gluon_resnext50_32x4d
In Collection: Gloun ResNeXt
Metadata:
FLOPs: 5472648192
Parameters: 25030000
File Size: 100441719
Architecture:
- 1x1 Convolution
- Batch Normalization
- Convolution
- Global Average Pooling
- Grouped Convolution
- Max Pooling
- ReLU
- ResNeXt Block
- Residual Connection
- Softmax
Tasks:
- Image Classification
Training Data:
- ImageNet
ID: gluon_resnext50_32x4d
Crop Pct: '0.875'
Image Size: '224'
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/gluon_resnet.py#L185
Weights: https://github.com/rwightman/pytorch-pretrained-gluonresnet/releases/download/v0.1/gluon_resnext50_32x4d-e6a097c1.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 79.35%
Top 5 Accuracy: 94.42%
-->
| pytorch-image-models/hfdocs/source/models/gloun-resnext.mdx/0 | {
"file_path": "pytorch-image-models/hfdocs/source/models/gloun-resnext.mdx",
"repo_id": "pytorch-image-models",
"token_count": 2709
} | 175 |
# NASNet
**NASNet** is a type of convolutional neural network discovered through neural architecture search. The building blocks consist of normal and reduction cells.
## How do I use this model on an image?
To load a pretrained model:
```py
>>> import timm
>>> model = timm.create_model('nasnetalarge', pretrained=True)
>>> model.eval()
```
To load and preprocess the image:
```py
>>> import urllib
>>> from PIL import Image
>>> from timm.data import resolve_data_config
>>> from timm.data.transforms_factory import create_transform
>>> config = resolve_data_config({}, model=model)
>>> transform = create_transform(**config)
>>> url, filename = ("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")
>>> urllib.request.urlretrieve(url, filename)
>>> img = Image.open(filename).convert('RGB')
>>> tensor = transform(img).unsqueeze(0) # transform and add batch dimension
```
To get the model predictions:
```py
>>> import torch
>>> with torch.no_grad():
... out = model(tensor)
>>> probabilities = torch.nn.functional.softmax(out[0], dim=0)
>>> print(probabilities.shape)
>>> # prints: torch.Size([1000])
```
To get the top-5 predictions class names:
```py
>>> # Get imagenet class mappings
>>> url, filename = ("https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt", "imagenet_classes.txt")
>>> urllib.request.urlretrieve(url, filename)
>>> with open("imagenet_classes.txt", "r") as f:
... categories = [s.strip() for s in f.readlines()]
>>> # Print top categories per image
>>> top5_prob, top5_catid = torch.topk(probabilities, 5)
>>> for i in range(top5_prob.size(0)):
... print(categories[top5_catid[i]], top5_prob[i].item())
>>> # prints class names and probabilities like:
>>> # [('Samoyed', 0.6425196528434753), ('Pomeranian', 0.04062102362513542), ('keeshond', 0.03186424449086189), ('white wolf', 0.01739676296710968), ('Eskimo dog', 0.011717947199940681)]
```
Replace the model name with the variant you want to use, e.g. `nasnetalarge`. You can find the IDs in the model summaries at the top of this page.
To extract image features with this model, follow the [timm feature extraction examples](../feature_extraction), just change the name of the model you want to use.
## How do I finetune this model?
You can finetune any of the pre-trained models just by changing the classifier (the last layer).
```py
>>> model = timm.create_model('nasnetalarge', pretrained=True, num_classes=NUM_FINETUNE_CLASSES)
```
To finetune on your own dataset, you have to write a training loop or adapt [timm's training
script](https://github.com/rwightman/pytorch-image-models/blob/master/train.py) to use your dataset.
## How do I train this model?
You can follow the [timm recipe scripts](../scripts) for training a new model afresh.
## Citation
```BibTeX
@misc{zoph2018learning,
title={Learning Transferable Architectures for Scalable Image Recognition},
author={Barret Zoph and Vijay Vasudevan and Jonathon Shlens and Quoc V. Le},
year={2018},
eprint={1707.07012},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
<!--
Type: model-index
Collections:
- Name: NASNet
Paper:
Title: Learning Transferable Architectures for Scalable Image Recognition
URL: https://paperswithcode.com/paper/learning-transferable-architectures-for
Models:
- Name: nasnetalarge
In Collection: NASNet
Metadata:
FLOPs: 30242402862
Parameters: 88750000
File Size: 356056626
Architecture:
- Average Pooling
- Batch Normalization
- Convolution
- Depthwise Separable Convolution
- Dropout
- ReLU
Tasks:
- Image Classification
Training Techniques:
- Label Smoothing
- RMSProp
- Weight Decay
Training Data:
- ImageNet
Training Resources: 50x Tesla K40 GPUs
ID: nasnetalarge
Dropout: 0.5
Crop Pct: '0.911'
Momentum: 0.9
Image Size: '331'
Interpolation: bicubic
Label Smoothing: 0.1
RMSProp \\( \epsilon \\): 1.0
Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/nasnet.py#L562
Weights: http://data.lip6.fr/cadene/pretrainedmodels/nasnetalarge-a1897284.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 82.63%
Top 5 Accuracy: 96.05%
-->
| pytorch-image-models/hfdocs/source/models/nasnet.mdx/0 | {
"file_path": "pytorch-image-models/hfdocs/source/models/nasnet.mdx",
"repo_id": "pytorch-image-models",
"token_count": 1536
} | 176 |
# SK-ResNeXt
**SK ResNeXt** is a variant of a [ResNeXt](https://www.paperswithcode.com/method/resnext) that employs a [Selective Kernel](https://paperswithcode.com/method/selective-kernel) unit. In general, all the large kernel convolutions in the original bottleneck blocks in ResNext are replaced by the proposed [SK convolutions](https://paperswithcode.com/method/selective-kernel-convolution), enabling the network to choose appropriate receptive field sizes in an adaptive manner.
## How do I use this model on an image?
To load a pretrained model:
```py
>>> import timm
>>> model = timm.create_model('skresnext50_32x4d', pretrained=True)
>>> model.eval()
```
To load and preprocess the image:
```py
>>> import urllib
>>> from PIL import Image
>>> from timm.data import resolve_data_config
>>> from timm.data.transforms_factory import create_transform
>>> config = resolve_data_config({}, model=model)
>>> transform = create_transform(**config)
>>> url, filename = ("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")
>>> urllib.request.urlretrieve(url, filename)
>>> img = Image.open(filename).convert('RGB')
>>> tensor = transform(img).unsqueeze(0) # transform and add batch dimension
```
To get the model predictions:
```py
>>> import torch
>>> with torch.no_grad():
... out = model(tensor)
>>> probabilities = torch.nn.functional.softmax(out[0], dim=0)
>>> print(probabilities.shape)
>>> # prints: torch.Size([1000])
```
To get the top-5 predictions class names:
```py
>>> # Get imagenet class mappings
>>> url, filename = ("https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt", "imagenet_classes.txt")
>>> urllib.request.urlretrieve(url, filename)
>>> with open("imagenet_classes.txt", "r") as f:
... categories = [s.strip() for s in f.readlines()]
>>> # Print top categories per image
>>> top5_prob, top5_catid = torch.topk(probabilities, 5)
>>> for i in range(top5_prob.size(0)):
... print(categories[top5_catid[i]], top5_prob[i].item())
>>> # prints class names and probabilities like:
>>> # [('Samoyed', 0.6425196528434753), ('Pomeranian', 0.04062102362513542), ('keeshond', 0.03186424449086189), ('white wolf', 0.01739676296710968), ('Eskimo dog', 0.011717947199940681)]
```
Replace the model name with the variant you want to use, e.g. `skresnext50_32x4d`. You can find the IDs in the model summaries at the top of this page.
To extract image features with this model, follow the [timm feature extraction examples](../feature_extraction), just change the name of the model you want to use.
## How do I finetune this model?
You can finetune any of the pre-trained models just by changing the classifier (the last layer).
```py
>>> model = timm.create_model('skresnext50_32x4d', pretrained=True, num_classes=NUM_FINETUNE_CLASSES)
```
To finetune on your own dataset, you have to write a training loop or adapt [timm's training
script](https://github.com/rwightman/pytorch-image-models/blob/master/train.py) to use your dataset.
## How do I train this model?
You can follow the [timm recipe scripts](../scripts) for training a new model afresh.
## Citation
```BibTeX
@misc{li2019selective,
title={Selective Kernel Networks},
author={Xiang Li and Wenhai Wang and Xiaolin Hu and Jian Yang},
year={2019},
eprint={1903.06586},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
<!--
Type: model-index
Collections:
- Name: SKResNeXt
Paper:
Title: Selective Kernel Networks
URL: https://paperswithcode.com/paper/selective-kernel-networks
Models:
- Name: skresnext50_32x4d
In Collection: SKResNeXt
Metadata:
FLOPs: 5739845824
Parameters: 27480000
File Size: 110340975
Architecture:
- Convolution
- Dense Connections
- Global Average Pooling
- Grouped Convolution
- Max Pooling
- Residual Connection
- Selective Kernel
- Softmax
Tasks:
- Image Classification
Training Data:
- ImageNet
Training Resources: 8x GPUs
ID: skresnext50_32x4d
LR: 0.1
Epochs: 100
Layers: 50
Crop Pct: '0.875'
Momentum: 0.9
Batch Size: 256
Image Size: '224'
Weight Decay: 0.0001
Interpolation: bicubic
Code: https://github.com/rwightman/pytorch-image-models/blob/a7f95818e44b281137503bcf4b3e3e94d8ffa52f/timm/models/sknet.py#L210
Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/skresnext50_ra-f40e40bf.pth
Results:
- Task: Image Classification
Dataset: ImageNet
Metrics:
Top 1 Accuracy: 80.15%
Top 5 Accuracy: 94.64%
--> | pytorch-image-models/hfdocs/source/models/skresnext.mdx/0 | {
"file_path": "pytorch-image-models/hfdocs/source/models/skresnext.mdx",
"repo_id": "pytorch-image-models",
"token_count": 1643
} | 177 |
# Models
[[autodoc]] timm.create_model
[[autodoc]] timm.list_models
| pytorch-image-models/hfdocs/source/reference/models.mdx/0 | {
"file_path": "pytorch-image-models/hfdocs/source/reference/models.mdx",
"repo_id": "pytorch-image-models",
"token_count": 29
} | 178 |
from abc import abstractmethod
class Reader:
def __init__(self):
pass
@abstractmethod
def _filename(self, index, basename=False, absolute=False):
pass
def filename(self, index, basename=False, absolute=False):
return self._filename(index, basename=basename, absolute=absolute)
def filenames(self, basename=False, absolute=False):
return [self._filename(index, basename=basename, absolute=absolute) for index in range(len(self))]
| pytorch-image-models/timm/data/readers/reader.py/0 | {
"file_path": "pytorch-image-models/timm/data/readers/reader.py",
"repo_id": "pytorch-image-models",
"token_count": 171
} | 179 |
""" Activations
A collection of jit-scripted activations fn and modules with a common interface so that they can
easily be swapped. All have an `inplace` arg even if not used.
All jit scripted activations are lacking in-place variations on purpose, scripted kernel fusion does not
currently work across in-place op boundaries, thus performance is equal to or less than the non-scripted
versions if they contain in-place ops.
Hacked together by / Copyright 2020 Ross Wightman
"""
import torch
from torch import nn as nn
from torch.nn import functional as F
@torch.jit.script
def swish_jit(x, inplace: bool = False):
"""Swish - Described in: https://arxiv.org/abs/1710.05941
"""
return x.mul(x.sigmoid())
@torch.jit.script
def mish_jit(x, _inplace: bool = False):
"""Mish: A Self Regularized Non-Monotonic Neural Activation Function - https://arxiv.org/abs/1908.08681
"""
return x.mul(F.softplus(x).tanh())
class SwishJit(nn.Module):
def __init__(self, inplace: bool = False):
super(SwishJit, self).__init__()
def forward(self, x):
return swish_jit(x)
class MishJit(nn.Module):
def __init__(self, inplace: bool = False):
super(MishJit, self).__init__()
def forward(self, x):
return mish_jit(x)
@torch.jit.script
def hard_sigmoid_jit(x, inplace: bool = False):
# return F.relu6(x + 3.) / 6.
return (x + 3).clamp(min=0, max=6).div(6.) # clamp seems ever so slightly faster?
class HardSigmoidJit(nn.Module):
def __init__(self, inplace: bool = False):
super(HardSigmoidJit, self).__init__()
def forward(self, x):
return hard_sigmoid_jit(x)
@torch.jit.script
def hard_swish_jit(x, inplace: bool = False):
# return x * (F.relu6(x + 3.) / 6)
return x * (x + 3).clamp(min=0, max=6).div(6.) # clamp seems ever so slightly faster?
class HardSwishJit(nn.Module):
def __init__(self, inplace: bool = False):
super(HardSwishJit, self).__init__()
def forward(self, x):
return hard_swish_jit(x)
@torch.jit.script
def hard_mish_jit(x, inplace: bool = False):
""" Hard Mish
Experimental, based on notes by Mish author Diganta Misra at
https://github.com/digantamisra98/H-Mish/blob/0da20d4bc58e696b6803f2523c58d3c8a82782d0/README.md
"""
return 0.5 * x * (x + 2).clamp(min=0, max=2)
class HardMishJit(nn.Module):
def __init__(self, inplace: bool = False):
super(HardMishJit, self).__init__()
def forward(self, x):
return hard_mish_jit(x)
| pytorch-image-models/timm/layers/activations_jit.py/0 | {
"file_path": "pytorch-image-models/timm/layers/activations_jit.py",
"repo_id": "pytorch-image-models",
"token_count": 1008
} | 180 |
""" Norm Layer Factory
Create norm modules by string (to mirror create_act and creat_norm-act fns)
Copyright 2022 Ross Wightman
"""
import functools
import types
from typing import Type
import torch.nn as nn
from .norm import GroupNorm, GroupNorm1, LayerNorm, LayerNorm2d, RmsNorm
from torchvision.ops.misc import FrozenBatchNorm2d
_NORM_MAP = dict(
batchnorm=nn.BatchNorm2d,
batchnorm2d=nn.BatchNorm2d,
batchnorm1d=nn.BatchNorm1d,
groupnorm=GroupNorm,
groupnorm1=GroupNorm1,
layernorm=LayerNorm,
layernorm2d=LayerNorm2d,
rmsnorm=RmsNorm,
frozenbatchnorm2d=FrozenBatchNorm2d,
)
_NORM_TYPES = {m for n, m in _NORM_MAP.items()}
def create_norm_layer(layer_name, num_features, **kwargs):
layer = get_norm_layer(layer_name)
layer_instance = layer(num_features, **kwargs)
return layer_instance
def get_norm_layer(norm_layer):
if norm_layer is None:
return None
assert isinstance(norm_layer, (type, str, types.FunctionType, functools.partial))
norm_kwargs = {}
# unbind partial fn, so args can be rebound later
if isinstance(norm_layer, functools.partial):
norm_kwargs.update(norm_layer.keywords)
norm_layer = norm_layer.func
if isinstance(norm_layer, str):
if not norm_layer:
return None
layer_name = norm_layer.replace('_', '')
norm_layer = _NORM_MAP[layer_name]
else:
norm_layer = norm_layer
if norm_kwargs:
norm_layer = functools.partial(norm_layer, **norm_kwargs) # bind/rebind args
return norm_layer
| pytorch-image-models/timm/layers/create_norm.py/0 | {
"file_path": "pytorch-image-models/timm/layers/create_norm.py",
"repo_id": "pytorch-image-models",
"token_count": 630
} | 181 |
""" Lambda Layer
Paper: `LambdaNetworks: Modeling Long-Range Interactions Without Attention`
- https://arxiv.org/abs/2102.08602
@misc{2102.08602,
Author = {Irwan Bello},
Title = {LambdaNetworks: Modeling Long-Range Interactions Without Attention},
Year = {2021},
}
Status:
This impl is a WIP. Code snippets in the paper were used as reference but
good chance some details are missing/wrong.
I've only implemented local lambda conv based pos embeddings.
For a PyTorch impl that includes other embedding options checkout
https://github.com/lucidrains/lambda-networks
Hacked together by / Copyright 2021 Ross Wightman
"""
import torch
from torch import nn
import torch.nn.functional as F
from .grid import ndgrid
from .helpers import to_2tuple, make_divisible
from .weight_init import trunc_normal_
def rel_pos_indices(size):
size = to_2tuple(size)
pos = torch.stack(ndgrid(torch.arange(size[0]), torch.arange(size[1]))).flatten(1)
rel_pos = pos[:, None, :] - pos[:, :, None]
rel_pos[0] += size[0] - 1
rel_pos[1] += size[1] - 1
return rel_pos # 2, H * W, H * W
class LambdaLayer(nn.Module):
"""Lambda Layer
Paper: `LambdaNetworks: Modeling Long-Range Interactions Without Attention`
- https://arxiv.org/abs/2102.08602
NOTE: intra-depth parameter 'u' is fixed at 1. It did not appear worth the complexity to add.
The internal dimensions of the lambda module are controlled via the interaction of several arguments.
* the output dimension of the module is specified by dim_out, which falls back to input dim if not set
* the value (v) dimension is set to dim_out // num_heads, the v projection determines the output dim
* the query (q) and key (k) dimension are determined by
* dim_head = (dim_out * attn_ratio // num_heads) if dim_head is None
* q = num_heads * dim_head, k = dim_head
* as seen above, attn_ratio determines the ratio of q and k relative to the output if dim_head not set
Args:
dim (int): input dimension to the module
dim_out (int): output dimension of the module, same as dim if not set
feat_size (Tuple[int, int]): size of input feature_map for relative pos variant H, W
stride (int): output stride of the module, avg pool used if stride == 2
num_heads (int): parallel attention heads.
dim_head (int): dimension of query and key heads, calculated from dim_out * attn_ratio // num_heads if not set
r (int): local lambda convolution radius. Use lambda conv if set, else relative pos if not. (default: 9)
qk_ratio (float): ratio of q and k dimensions to output dimension when dim_head not set. (default: 1.0)
qkv_bias (bool): add bias to q, k, and v projections
"""
def __init__(
self, dim, dim_out=None, feat_size=None, stride=1, num_heads=4, dim_head=16, r=9,
qk_ratio=1.0, qkv_bias=False):
super().__init__()
dim_out = dim_out or dim
assert dim_out % num_heads == 0, ' should be divided by num_heads'
self.dim_qk = dim_head or make_divisible(dim_out * qk_ratio, divisor=8) // num_heads
self.num_heads = num_heads
self.dim_v = dim_out // num_heads
self.qkv = nn.Conv2d(
dim,
num_heads * self.dim_qk + self.dim_qk + self.dim_v,
kernel_size=1, bias=qkv_bias)
self.norm_q = nn.BatchNorm2d(num_heads * self.dim_qk)
self.norm_v = nn.BatchNorm2d(self.dim_v)
if r is not None:
# local lambda convolution for pos
self.conv_lambda = nn.Conv3d(1, self.dim_qk, (r, r, 1), padding=(r // 2, r // 2, 0))
self.pos_emb = None
self.rel_pos_indices = None
else:
# relative pos embedding
assert feat_size is not None
feat_size = to_2tuple(feat_size)
rel_size = [2 * s - 1 for s in feat_size]
self.conv_lambda = None
self.pos_emb = nn.Parameter(torch.zeros(rel_size[0], rel_size[1], self.dim_qk))
self.register_buffer('rel_pos_indices', rel_pos_indices(feat_size), persistent=False)
self.pool = nn.AvgPool2d(2, 2) if stride == 2 else nn.Identity()
self.reset_parameters()
def reset_parameters(self):
trunc_normal_(self.qkv.weight, std=self.qkv.weight.shape[1] ** -0.5) # fan-in
if self.conv_lambda is not None:
trunc_normal_(self.conv_lambda.weight, std=self.dim_qk ** -0.5)
if self.pos_emb is not None:
trunc_normal_(self.pos_emb, std=.02)
def forward(self, x):
B, C, H, W = x.shape
M = H * W
qkv = self.qkv(x)
q, k, v = torch.split(qkv, [
self.num_heads * self.dim_qk, self.dim_qk, self.dim_v], dim=1)
q = self.norm_q(q).reshape(B, self.num_heads, self.dim_qk, M).transpose(-1, -2) # B, num_heads, M, K
v = self.norm_v(v).reshape(B, self.dim_v, M).transpose(-1, -2) # B, M, V
k = F.softmax(k.reshape(B, self.dim_qk, M), dim=-1) # B, K, M
content_lam = k @ v # B, K, V
content_out = q @ content_lam.unsqueeze(1) # B, num_heads, M, V
if self.pos_emb is None:
position_lam = self.conv_lambda(v.reshape(B, 1, H, W, self.dim_v)) # B, H, W, V, K
position_lam = position_lam.reshape(B, 1, self.dim_qk, H * W, self.dim_v).transpose(2, 3) # B, 1, M, K, V
else:
# FIXME relative pos embedding path not fully verified
pos_emb = self.pos_emb[self.rel_pos_indices[0], self.rel_pos_indices[1]].expand(B, -1, -1, -1)
position_lam = (pos_emb.transpose(-1, -2) @ v.unsqueeze(1)).unsqueeze(1) # B, 1, M, K, V
position_out = (q.unsqueeze(-2) @ position_lam).squeeze(-2) # B, num_heads, M, V
out = (content_out + position_out).transpose(-1, -2).reshape(B, C, H, W) # B, C (num_heads * V), H, W
out = self.pool(out)
return out
| pytorch-image-models/timm/layers/lambda_layer.py/0 | {
"file_path": "pytorch-image-models/timm/layers/lambda_layer.py",
"repo_id": "pytorch-image-models",
"token_count": 2611
} | 182 |
""" Selective Kernel Convolution/Attention
Paper: Selective Kernel Networks (https://arxiv.org/abs/1903.06586)
Hacked together by / Copyright 2020 Ross Wightman
"""
import torch
from torch import nn as nn
from .conv_bn_act import ConvNormActAa
from .helpers import make_divisible
from .trace_utils import _assert
def _kernel_valid(k):
if isinstance(k, (list, tuple)):
for ki in k:
return _kernel_valid(ki)
assert k >= 3 and k % 2
class SelectiveKernelAttn(nn.Module):
def __init__(self, channels, num_paths=2, attn_channels=32, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d):
""" Selective Kernel Attention Module
Selective Kernel attention mechanism factored out into its own module.
"""
super(SelectiveKernelAttn, self).__init__()
self.num_paths = num_paths
self.fc_reduce = nn.Conv2d(channels, attn_channels, kernel_size=1, bias=False)
self.bn = norm_layer(attn_channels)
self.act = act_layer(inplace=True)
self.fc_select = nn.Conv2d(attn_channels, channels * num_paths, kernel_size=1, bias=False)
def forward(self, x):
_assert(x.shape[1] == self.num_paths, '')
x = x.sum(1).mean((2, 3), keepdim=True)
x = self.fc_reduce(x)
x = self.bn(x)
x = self.act(x)
x = self.fc_select(x)
B, C, H, W = x.shape
x = x.view(B, self.num_paths, C // self.num_paths, H, W)
x = torch.softmax(x, dim=1)
return x
class SelectiveKernel(nn.Module):
def __init__(self, in_channels, out_channels=None, kernel_size=None, stride=1, dilation=1, groups=1,
rd_ratio=1./16, rd_channels=None, rd_divisor=8, keep_3x3=True, split_input=True,
act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d, aa_layer=None, drop_layer=None):
""" Selective Kernel Convolution Module
As described in Selective Kernel Networks (https://arxiv.org/abs/1903.06586) with some modifications.
Largest change is the input split, which divides the input channels across each convolution path, this can
be viewed as a grouping of sorts, but the output channel counts expand to the module level value. This keeps
the parameter count from ballooning when the convolutions themselves don't have groups, but still provides
a noteworthy increase in performance over similar param count models without this attention layer. -Ross W
Args:
in_channels (int): module input (feature) channel count
out_channels (int): module output (feature) channel count
kernel_size (int, list): kernel size for each convolution branch
stride (int): stride for convolutions
dilation (int): dilation for module as a whole, impacts dilation of each branch
groups (int): number of groups for each branch
rd_ratio (int, float): reduction factor for attention features
keep_3x3 (bool): keep all branch convolution kernels as 3x3, changing larger kernels for dilations
split_input (bool): split input channels evenly across each convolution branch, keeps param count lower,
can be viewed as grouping by path, output expands to module out_channels count
act_layer (nn.Module): activation layer to use
norm_layer (nn.Module): batchnorm/norm layer to use
aa_layer (nn.Module): anti-aliasing module
drop_layer (nn.Module): spatial drop module in convs (drop block, etc)
"""
super(SelectiveKernel, self).__init__()
out_channels = out_channels or in_channels
kernel_size = kernel_size or [3, 5] # default to one 3x3 and one 5x5 branch. 5x5 -> 3x3 + dilation
_kernel_valid(kernel_size)
if not isinstance(kernel_size, list):
kernel_size = [kernel_size] * 2
if keep_3x3:
dilation = [dilation * (k - 1) // 2 for k in kernel_size]
kernel_size = [3] * len(kernel_size)
else:
dilation = [dilation] * len(kernel_size)
self.num_paths = len(kernel_size)
self.in_channels = in_channels
self.out_channels = out_channels
self.split_input = split_input
if self.split_input:
assert in_channels % self.num_paths == 0
in_channels = in_channels // self.num_paths
groups = min(out_channels, groups)
conv_kwargs = dict(
stride=stride, groups=groups, act_layer=act_layer, norm_layer=norm_layer,
aa_layer=aa_layer, drop_layer=drop_layer)
self.paths = nn.ModuleList([
ConvNormActAa(in_channels, out_channels, kernel_size=k, dilation=d, **conv_kwargs)
for k, d in zip(kernel_size, dilation)])
attn_channels = rd_channels or make_divisible(out_channels * rd_ratio, divisor=rd_divisor)
self.attn = SelectiveKernelAttn(out_channels, self.num_paths, attn_channels)
def forward(self, x):
if self.split_input:
x_split = torch.split(x, self.in_channels // self.num_paths, 1)
x_paths = [op(x_split[i]) for i, op in enumerate(self.paths)]
else:
x_paths = [op(x) for op in self.paths]
x = torch.stack(x_paths, dim=1)
x_attn = self.attn(x)
x = x * x_attn
x = torch.sum(x, dim=1)
return x
| pytorch-image-models/timm/layers/selective_kernel.py/0 | {
"file_path": "pytorch-image-models/timm/layers/selective_kernel.py",
"repo_id": "pytorch-image-models",
"token_count": 2318
} | 183 |
from .beit import *
from .byoanet import *
from .byobnet import *
from .cait import *
from .coat import *
from .convit import *
from .convmixer import *
from .convnext import *
from .crossvit import *
from .cspnet import *
from .davit import *
from .deit import *
from .densenet import *
from .dla import *
from .dpn import *
from .edgenext import *
from .efficientformer import *
from .efficientformer_v2 import *
from .efficientnet import *
from .efficientvit_mit import *
from .efficientvit_msra import *
from .eva import *
from .fastvit import *
from .focalnet import *
from .gcvit import *
from .ghostnet import *
from .hardcorenas import *
from .hrnet import *
from .inception_next import *
from .inception_resnet_v2 import *
from .inception_v3 import *
from .inception_v4 import *
from .levit import *
from .maxxvit import *
from .metaformer import *
from .mlp_mixer import *
from .mobilenetv3 import *
from .mobilevit import *
from .mvitv2 import *
from .nasnet import *
from .nest import *
from .nfnet import *
from .pit import *
from .pnasnet import *
from .pvt_v2 import *
from .regnet import *
from .repghost import *
from .repvit import *
from .res2net import *
from .resnest import *
from .resnet import *
from .resnetv2 import *
from .rexnet import *
from .selecsls import *
from .senet import *
from .sequencer import *
from .sknet import *
from .swin_transformer import *
from .swin_transformer_v2 import *
from .swin_transformer_v2_cr import *
from .tiny_vit import *
from .tnt import *
from .tresnet import *
from .twins import *
from .vgg import *
from .visformer import *
from .vision_transformer import *
from .vision_transformer_hybrid import *
from .vision_transformer_relpos import *
from .vision_transformer_sam import *
from .volo import *
from .vovnet import *
from .xception import *
from .xception_aligned import *
from .xcit import *
from ._builder import build_model_with_cfg, load_pretrained, load_custom_pretrained, resolve_pretrained_cfg, \
set_pretrained_download_progress, set_pretrained_check_hash
from ._factory import create_model, parse_model_name, safe_model_name
from ._features import FeatureInfo, FeatureHooks, FeatureHookNet, FeatureListNet, FeatureDictNet
from ._features_fx import FeatureGraphNet, GraphExtractNet, create_feature_extractor, \
register_notrace_module, is_notrace_module, get_notrace_modules, \
register_notrace_function, is_notrace_function, get_notrace_functions
from ._helpers import clean_state_dict, load_state_dict, load_checkpoint, remap_state_dict, resume_checkpoint
from ._hub import load_model_config_from_hf, load_state_dict_from_hf, push_to_hf_hub
from ._manipulate import model_parameters, named_apply, named_modules, named_modules_with_params, \
group_modules, group_parameters, checkpoint_seq, adapt_input_conv
from ._pretrained import PretrainedCfg, DefaultCfg, filter_pretrained_cfg
from ._prune import adapt_model_from_string
from ._registry import split_model_name_tag, get_arch_name, generate_default_cfgs, register_model, \
register_model_deprecations, model_entrypoint, list_models, list_pretrained, get_deprecated_models, \
is_model, list_modules, is_model_in_modules, is_model_pretrained, get_pretrained_cfg, get_pretrained_cfg_value
| pytorch-image-models/timm/models/__init__.py/0 | {
"file_path": "pytorch-image-models/timm/models/__init__.py",
"repo_id": "pytorch-image-models",
"token_count": 1061
} | 184 |
""" PyTorch implementation of DualPathNetworks
Based on original MXNet implementation https://github.com/cypw/DPNs with
many ideas from another PyTorch implementation https://github.com/oyam/pytorch-DPNs.
This implementation is compatible with the pretrained weights from cypw's MXNet implementation.
Hacked together by / Copyright 2020 Ross Wightman
"""
from collections import OrderedDict
from functools import partial
from typing import Tuple
import torch
import torch.nn as nn
import torch.nn.functional as F
from timm.data import IMAGENET_DPN_MEAN, IMAGENET_DPN_STD, IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
from timm.layers import BatchNormAct2d, ConvNormAct, create_conv2d, create_classifier, get_norm_act_layer
from ._builder import build_model_with_cfg
from ._registry import register_model, generate_default_cfgs
__all__ = ['DPN']
class CatBnAct(nn.Module):
def __init__(self, in_chs, norm_layer=BatchNormAct2d):
super(CatBnAct, self).__init__()
self.bn = norm_layer(in_chs, eps=0.001)
@torch.jit._overload_method # noqa: F811
def forward(self, x):
# type: (Tuple[torch.Tensor, torch.Tensor]) -> (torch.Tensor)
pass
@torch.jit._overload_method # noqa: F811
def forward(self, x):
# type: (torch.Tensor) -> (torch.Tensor)
pass
def forward(self, x):
if isinstance(x, tuple):
x = torch.cat(x, dim=1)
return self.bn(x)
class BnActConv2d(nn.Module):
def __init__(self, in_chs, out_chs, kernel_size, stride, groups=1, norm_layer=BatchNormAct2d):
super(BnActConv2d, self).__init__()
self.bn = norm_layer(in_chs, eps=0.001)
self.conv = create_conv2d(in_chs, out_chs, kernel_size, stride=stride, groups=groups)
def forward(self, x):
return self.conv(self.bn(x))
class DualPathBlock(nn.Module):
def __init__(
self,
in_chs,
num_1x1_a,
num_3x3_b,
num_1x1_c,
inc,
groups,
block_type='normal',
b=False,
):
super(DualPathBlock, self).__init__()
self.num_1x1_c = num_1x1_c
self.inc = inc
self.b = b
if block_type == 'proj':
self.key_stride = 1
self.has_proj = True
elif block_type == 'down':
self.key_stride = 2
self.has_proj = True
else:
assert block_type == 'normal'
self.key_stride = 1
self.has_proj = False
self.c1x1_w_s1 = None
self.c1x1_w_s2 = None
if self.has_proj:
# Using different member names here to allow easier parameter key matching for conversion
if self.key_stride == 2:
self.c1x1_w_s2 = BnActConv2d(
in_chs=in_chs, out_chs=num_1x1_c + 2 * inc, kernel_size=1, stride=2)
else:
self.c1x1_w_s1 = BnActConv2d(
in_chs=in_chs, out_chs=num_1x1_c + 2 * inc, kernel_size=1, stride=1)
self.c1x1_a = BnActConv2d(in_chs=in_chs, out_chs=num_1x1_a, kernel_size=1, stride=1)
self.c3x3_b = BnActConv2d(
in_chs=num_1x1_a, out_chs=num_3x3_b, kernel_size=3, stride=self.key_stride, groups=groups)
if b:
self.c1x1_c = CatBnAct(in_chs=num_3x3_b)
self.c1x1_c1 = create_conv2d(num_3x3_b, num_1x1_c, kernel_size=1)
self.c1x1_c2 = create_conv2d(num_3x3_b, inc, kernel_size=1)
else:
self.c1x1_c = BnActConv2d(in_chs=num_3x3_b, out_chs=num_1x1_c + inc, kernel_size=1, stride=1)
self.c1x1_c1 = None
self.c1x1_c2 = None
@torch.jit._overload_method # noqa: F811
def forward(self, x):
# type: (Tuple[torch.Tensor, torch.Tensor]) -> Tuple[torch.Tensor, torch.Tensor]
pass
@torch.jit._overload_method # noqa: F811
def forward(self, x):
# type: (torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]
pass
def forward(self, x) -> Tuple[torch.Tensor, torch.Tensor]:
if isinstance(x, tuple):
x_in = torch.cat(x, dim=1)
else:
x_in = x
if self.c1x1_w_s1 is None and self.c1x1_w_s2 is None:
# self.has_proj == False, torchscript requires condition on module == None
x_s1 = x[0]
x_s2 = x[1]
else:
# self.has_proj == True
if self.c1x1_w_s1 is not None:
# self.key_stride = 1
x_s = self.c1x1_w_s1(x_in)
else:
# self.key_stride = 2
x_s = self.c1x1_w_s2(x_in)
x_s1 = x_s[:, :self.num_1x1_c, :, :]
x_s2 = x_s[:, self.num_1x1_c:, :, :]
x_in = self.c1x1_a(x_in)
x_in = self.c3x3_b(x_in)
x_in = self.c1x1_c(x_in)
if self.c1x1_c1 is not None:
# self.b == True, using None check for torchscript compat
out1 = self.c1x1_c1(x_in)
out2 = self.c1x1_c2(x_in)
else:
out1 = x_in[:, :self.num_1x1_c, :, :]
out2 = x_in[:, self.num_1x1_c:, :, :]
resid = x_s1 + out1
dense = torch.cat([x_s2, out2], dim=1)
return resid, dense
class DPN(nn.Module):
def __init__(
self,
k_sec=(3, 4, 20, 3),
inc_sec=(16, 32, 24, 128),
k_r=96,
groups=32,
num_classes=1000,
in_chans=3,
output_stride=32,
global_pool='avg',
small=False,
num_init_features=64,
b=False,
drop_rate=0.,
norm_layer='batchnorm2d',
act_layer='relu',
fc_act_layer='elu',
):
super(DPN, self).__init__()
self.num_classes = num_classes
self.drop_rate = drop_rate
self.b = b
assert output_stride == 32 # FIXME look into dilation support
norm_layer = partial(get_norm_act_layer(norm_layer, act_layer=act_layer), eps=.001)
fc_norm_layer = partial(get_norm_act_layer(norm_layer, act_layer=fc_act_layer), eps=.001, inplace=False)
bw_factor = 1 if small else 4
blocks = OrderedDict()
# conv1
blocks['conv1_1'] = ConvNormAct(
in_chans, num_init_features, kernel_size=3 if small else 7, stride=2, norm_layer=norm_layer)
blocks['conv1_pool'] = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.feature_info = [dict(num_chs=num_init_features, reduction=2, module='features.conv1_1')]
# conv2
bw = 64 * bw_factor
inc = inc_sec[0]
r = (k_r * bw) // (64 * bw_factor)
blocks['conv2_1'] = DualPathBlock(num_init_features, r, r, bw, inc, groups, 'proj', b)
in_chs = bw + 3 * inc
for i in range(2, k_sec[0] + 1):
blocks['conv2_' + str(i)] = DualPathBlock(in_chs, r, r, bw, inc, groups, 'normal', b)
in_chs += inc
self.feature_info += [dict(num_chs=in_chs, reduction=4, module=f'features.conv2_{k_sec[0]}')]
# conv3
bw = 128 * bw_factor
inc = inc_sec[1]
r = (k_r * bw) // (64 * bw_factor)
blocks['conv3_1'] = DualPathBlock(in_chs, r, r, bw, inc, groups, 'down', b)
in_chs = bw + 3 * inc
for i in range(2, k_sec[1] + 1):
blocks['conv3_' + str(i)] = DualPathBlock(in_chs, r, r, bw, inc, groups, 'normal', b)
in_chs += inc
self.feature_info += [dict(num_chs=in_chs, reduction=8, module=f'features.conv3_{k_sec[1]}')]
# conv4
bw = 256 * bw_factor
inc = inc_sec[2]
r = (k_r * bw) // (64 * bw_factor)
blocks['conv4_1'] = DualPathBlock(in_chs, r, r, bw, inc, groups, 'down', b)
in_chs = bw + 3 * inc
for i in range(2, k_sec[2] + 1):
blocks['conv4_' + str(i)] = DualPathBlock(in_chs, r, r, bw, inc, groups, 'normal', b)
in_chs += inc
self.feature_info += [dict(num_chs=in_chs, reduction=16, module=f'features.conv4_{k_sec[2]}')]
# conv5
bw = 512 * bw_factor
inc = inc_sec[3]
r = (k_r * bw) // (64 * bw_factor)
blocks['conv5_1'] = DualPathBlock(in_chs, r, r, bw, inc, groups, 'down', b)
in_chs = bw + 3 * inc
for i in range(2, k_sec[3] + 1):
blocks['conv5_' + str(i)] = DualPathBlock(in_chs, r, r, bw, inc, groups, 'normal', b)
in_chs += inc
self.feature_info += [dict(num_chs=in_chs, reduction=32, module=f'features.conv5_{k_sec[3]}')]
blocks['conv5_bn_ac'] = CatBnAct(in_chs, norm_layer=fc_norm_layer)
self.num_features = in_chs
self.features = nn.Sequential(blocks)
# Using 1x1 conv for the FC layer to allow the extra pooling scheme
self.global_pool, self.classifier = create_classifier(
self.num_features, self.num_classes, pool_type=global_pool, use_conv=True)
self.flatten = nn.Flatten(1) if global_pool else nn.Identity()
@torch.jit.ignore
def group_matcher(self, coarse=False):
matcher = dict(
stem=r'^features\.conv1',
blocks=[
(r'^features\.conv(\d+)' if coarse else r'^features\.conv(\d+)_(\d+)', None),
(r'^features\.conv5_bn_ac', (99999,))
]
)
return matcher
@torch.jit.ignore
def set_grad_checkpointing(self, enable=True):
assert not enable, 'gradient checkpointing not supported'
@torch.jit.ignore
def get_classifier(self):
return self.classifier
def reset_classifier(self, num_classes, global_pool='avg'):
self.num_classes = num_classes
self.global_pool, self.classifier = create_classifier(
self.num_features, self.num_classes, pool_type=global_pool, use_conv=True)
self.flatten = nn.Flatten(1) if global_pool else nn.Identity()
def forward_features(self, x):
return self.features(x)
def forward_head(self, x, pre_logits: bool = False):
x = self.global_pool(x)
if self.drop_rate > 0.:
x = F.dropout(x, p=self.drop_rate, training=self.training)
if pre_logits:
return self.flatten(x)
x = self.classifier(x)
return self.flatten(x)
def forward(self, x):
x = self.forward_features(x)
x = self.forward_head(x)
return x
def _create_dpn(variant, pretrained=False, **kwargs):
return build_model_with_cfg(
DPN,
variant,
pretrained,
feature_cfg=dict(feature_concat=True, flatten_sequential=True),
**kwargs,
)
def _cfg(url='', **kwargs):
return {
'url': url, 'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': (7, 7),
'crop_pct': 0.875, 'interpolation': 'bicubic',
'mean': IMAGENET_DPN_MEAN, 'std': IMAGENET_DPN_STD,
'first_conv': 'features.conv1_1.conv', 'classifier': 'classifier',
**kwargs
}
default_cfgs = generate_default_cfgs({
'dpn48b.untrained': _cfg(mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD),
'dpn68.mx_in1k': _cfg(hf_hub_id='timm/'),
'dpn68b.ra_in1k': _cfg(
hf_hub_id='timm/',
mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD,
crop_pct=0.95, test_input_size=(3, 288, 288), test_crop_pct=1.0),
'dpn68b.mx_in1k': _cfg(hf_hub_id='timm/'),
'dpn92.mx_in1k': _cfg(hf_hub_id='timm/'),
'dpn98.mx_in1k': _cfg(hf_hub_id='timm/'),
'dpn131.mx_in1k': _cfg(hf_hub_id='timm/'),
'dpn107.mx_in1k': _cfg(hf_hub_id='timm/')
})
@register_model
def dpn48b(pretrained=False, **kwargs) -> DPN:
model_args = dict(
small=True, num_init_features=10, k_r=128, groups=32,
b=True, k_sec=(3, 4, 6, 3), inc_sec=(16, 32, 32, 64), act_layer='silu')
return _create_dpn('dpn48b', pretrained=pretrained, **dict(model_args, **kwargs))
@register_model
def dpn68(pretrained=False, **kwargs) -> DPN:
model_args = dict(
small=True, num_init_features=10, k_r=128, groups=32,
k_sec=(3, 4, 12, 3), inc_sec=(16, 32, 32, 64))
return _create_dpn('dpn68', pretrained=pretrained, **dict(model_args, **kwargs))
@register_model
def dpn68b(pretrained=False, **kwargs) -> DPN:
model_args = dict(
small=True, num_init_features=10, k_r=128, groups=32,
b=True, k_sec=(3, 4, 12, 3), inc_sec=(16, 32, 32, 64))
return _create_dpn('dpn68b', pretrained=pretrained, **dict(model_args, **kwargs))
@register_model
def dpn92(pretrained=False, **kwargs) -> DPN:
model_args = dict(
num_init_features=64, k_r=96, groups=32,
k_sec=(3, 4, 20, 3), inc_sec=(16, 32, 24, 128))
return _create_dpn('dpn92', pretrained=pretrained, **dict(model_args, **kwargs))
@register_model
def dpn98(pretrained=False, **kwargs) -> DPN:
model_args = dict(
num_init_features=96, k_r=160, groups=40,
k_sec=(3, 6, 20, 3), inc_sec=(16, 32, 32, 128))
return _create_dpn('dpn98', pretrained=pretrained, **dict(model_args, **kwargs))
@register_model
def dpn131(pretrained=False, **kwargs) -> DPN:
model_args = dict(
num_init_features=128, k_r=160, groups=40,
k_sec=(4, 8, 28, 3), inc_sec=(16, 32, 32, 128))
return _create_dpn('dpn131', pretrained=pretrained, **dict(model_args, **kwargs))
@register_model
def dpn107(pretrained=False, **kwargs) -> DPN:
model_args = dict(
num_init_features=128, k_r=200, groups=50,
k_sec=(4, 8, 20, 3), inc_sec=(20, 64, 64, 128))
return _create_dpn('dpn107', pretrained=pretrained, **dict(model_args, **kwargs))
| pytorch-image-models/timm/models/dpn.py/0 | {
"file_path": "pytorch-image-models/timm/models/dpn.py",
"repo_id": "pytorch-image-models",
"token_count": 6985
} | 185 |
from ._builder import *
from ._helpers import *
from ._manipulate import *
from ._prune import *
import warnings
warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.models", DeprecationWarning)
| pytorch-image-models/timm/models/helpers.py/0 | {
"file_path": "pytorch-image-models/timm/models/helpers.py",
"repo_id": "pytorch-image-models",
"token_count": 64
} | 186 |
""" Nested Transformer (NesT) in PyTorch
A PyTorch implement of Aggregating Nested Transformers as described in:
'Aggregating Nested Transformers'
- https://arxiv.org/abs/2105.12723
The official Jax code is released and available at https://github.com/google-research/nested-transformer. The weights
have been converted with convert/convert_nest_flax.py
Acknowledgments:
* The paper authors for sharing their research, code, and model weights
* Ross Wightman's existing code off which I based this
Copyright 2021 Alexander Soare
"""
import collections.abc
import logging
import math
from functools import partial
import torch
import torch.nn.functional as F
from torch import nn
from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
from timm.layers import PatchEmbed, Mlp, DropPath, create_classifier, trunc_normal_, _assert
from timm.layers import create_conv2d, create_pool2d, to_ntuple, use_fused_attn, LayerNorm
from ._builder import build_model_with_cfg
from ._features_fx import register_notrace_function
from ._manipulate import checkpoint_seq, named_apply
from ._registry import register_model, generate_default_cfgs, register_model_deprecations
__all__ = ['Nest'] # model_registry will add each entrypoint fn to this
_logger = logging.getLogger(__name__)
class Attention(nn.Module):
"""
This is much like `.vision_transformer.Attention` but uses *localised* self attention by accepting an input with
an extra "image block" dim
"""
fused_attn: torch.jit.Final[bool]
def __init__(self, dim, num_heads=8, qkv_bias=False, attn_drop=0., proj_drop=0.):
super().__init__()
self.num_heads = num_heads
head_dim = dim // num_heads
self.scale = head_dim ** -0.5
self.fused_attn = use_fused_attn()
self.qkv = nn.Linear(dim, 3*dim, bias=qkv_bias)
self.attn_drop = nn.Dropout(attn_drop)
self.proj = nn.Linear(dim, dim)
self.proj_drop = nn.Dropout(proj_drop)
def forward(self, x):
"""
x is shape: B (batch_size), T (image blocks), N (seq length per image block), C (embed dim)
"""
B, T, N, C = x.shape
# result of next line is (qkv, B, num (H)eads, T, N, (C')hannels per head)
qkv = self.qkv(x).reshape(B, T, N, 3, self.num_heads, C // self.num_heads).permute(3, 0, 4, 1, 2, 5)
q, k, v = qkv.unbind(0) # make torchscript happy (cannot use tensor as tuple)
if self.fused_attn:
x = F.scaled_dot_product_attention(q, k, v, dropout_p=self.attn_drop.p if self.training else 0.)
else:
q = q * self.scale
attn = q @ k.transpose(-2, -1) # (B, H, T, N, N)
attn = attn.softmax(dim=-1)
attn = self.attn_drop(attn)
x = attn @ v
# (B, H, T, N, C'), permute -> (B, T, N, C', H)
x = x.permute(0, 2, 3, 4, 1).reshape(B, T, N, C)
x = self.proj(x)
x = self.proj_drop(x)
return x # (B, T, N, C)
class TransformerLayer(nn.Module):
"""
This is much like `.vision_transformer.Block` but:
- Called TransformerLayer here to allow for "block" as defined in the paper ("non-overlapping image blocks")
- Uses modified Attention layer that handles the "block" dimension
"""
def __init__(
self,
dim,
num_heads,
mlp_ratio=4.,
qkv_bias=False,
proj_drop=0.,
attn_drop=0.,
drop_path=0.,
act_layer=nn.GELU,
norm_layer=nn.LayerNorm,
):
super().__init__()
self.norm1 = norm_layer(dim)
self.attn = Attention(
dim,
num_heads=num_heads,
qkv_bias=qkv_bias,
attn_drop=attn_drop,
proj_drop=proj_drop,
)
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
self.norm2 = norm_layer(dim)
mlp_hidden_dim = int(dim * mlp_ratio)
self.mlp = Mlp(
in_features=dim,
hidden_features=mlp_hidden_dim,
act_layer=act_layer,
drop=proj_drop,
)
def forward(self, x):
y = self.norm1(x)
x = x + self.drop_path(self.attn(y))
x = x + self.drop_path(self.mlp(self.norm2(x)))
return x
class ConvPool(nn.Module):
def __init__(self, in_channels, out_channels, norm_layer, pad_type=''):
super().__init__()
self.conv = create_conv2d(in_channels, out_channels, kernel_size=3, padding=pad_type, bias=True)
self.norm = norm_layer(out_channels)
self.pool = create_pool2d('max', kernel_size=3, stride=2, padding=pad_type)
def forward(self, x):
"""
x is expected to have shape (B, C, H, W)
"""
_assert(x.shape[-2] % 2 == 0, 'BlockAggregation requires even input spatial dims')
_assert(x.shape[-1] % 2 == 0, 'BlockAggregation requires even input spatial dims')
x = self.conv(x)
# Layer norm done over channel dim only
x = self.norm(x.permute(0, 2, 3, 1)).permute(0, 3, 1, 2)
x = self.pool(x)
return x # (B, C, H//2, W//2)
def blockify(x, block_size: int):
"""image to blocks
Args:
x (Tensor): with shape (B, H, W, C)
block_size (int): edge length of a single square block in units of H, W
"""
B, H, W, C = x.shape
_assert(H % block_size == 0, '`block_size` must divide input height evenly')
_assert(W % block_size == 0, '`block_size` must divide input width evenly')
grid_height = H // block_size
grid_width = W // block_size
x = x.reshape(B, grid_height, block_size, grid_width, block_size, C)
x = x.transpose(2, 3).reshape(B, grid_height * grid_width, -1, C)
return x # (B, T, N, C)
@register_notrace_function # reason: int receives Proxy
def deblockify(x, block_size: int):
"""blocks to image
Args:
x (Tensor): with shape (B, T, N, C) where T is number of blocks and N is sequence size per block
block_size (int): edge length of a single square block in units of desired H, W
"""
B, T, _, C = x.shape
grid_size = int(math.sqrt(T))
height = width = grid_size * block_size
x = x.reshape(B, grid_size, grid_size, block_size, block_size, C)
x = x.transpose(2, 3).reshape(B, height, width, C)
return x # (B, H, W, C)
class NestLevel(nn.Module):
""" Single hierarchical level of a Nested Transformer
"""
def __init__(
self,
num_blocks,
block_size,
seq_length,
num_heads,
depth,
embed_dim,
prev_embed_dim=None,
mlp_ratio=4.,
qkv_bias=True,
proj_drop=0.,
attn_drop=0.,
drop_path=[],
norm_layer=None,
act_layer=None,
pad_type='',
):
super().__init__()
self.block_size = block_size
self.grad_checkpointing = False
self.pos_embed = nn.Parameter(torch.zeros(1, num_blocks, seq_length, embed_dim))
if prev_embed_dim is not None:
self.pool = ConvPool(prev_embed_dim, embed_dim, norm_layer=norm_layer, pad_type=pad_type)
else:
self.pool = nn.Identity()
# Transformer encoder
if len(drop_path):
assert len(drop_path) == depth, 'Must provide as many drop path rates as there are transformer layers'
self.transformer_encoder = nn.Sequential(*[
TransformerLayer(
dim=embed_dim,
num_heads=num_heads,
mlp_ratio=mlp_ratio,
qkv_bias=qkv_bias,
proj_drop=proj_drop,
attn_drop=attn_drop,
drop_path=drop_path[i],
norm_layer=norm_layer,
act_layer=act_layer,
)
for i in range(depth)])
def forward(self, x):
"""
expects x as (B, C, H, W)
"""
x = self.pool(x)
x = x.permute(0, 2, 3, 1) # (B, H', W', C), switch to channels last for transformer
x = blockify(x, self.block_size) # (B, T, N, C')
x = x + self.pos_embed
if self.grad_checkpointing and not torch.jit.is_scripting():
x = checkpoint_seq(self.transformer_encoder, x)
else:
x = self.transformer_encoder(x) # (B, T, N, C')
x = deblockify(x, self.block_size) # (B, H', W', C')
# Channel-first for block aggregation, and generally to replicate convnet feature map at each stage
return x.permute(0, 3, 1, 2) # (B, C, H', W')
class Nest(nn.Module):
""" Nested Transformer (NesT)
A PyTorch impl of : `Aggregating Nested Transformers`
- https://arxiv.org/abs/2105.12723
"""
def __init__(
self,
img_size=224,
in_chans=3,
patch_size=4,
num_levels=3,
embed_dims=(128, 256, 512),
num_heads=(4, 8, 16),
depths=(2, 2, 20),
num_classes=1000,
mlp_ratio=4.,
qkv_bias=True,
drop_rate=0.,
proj_drop_rate=0.,
attn_drop_rate=0.,
drop_path_rate=0.5,
norm_layer=None,
act_layer=None,
pad_type='',
weight_init='',
global_pool='avg',
):
"""
Args:
img_size (int, tuple): input image size
in_chans (int): number of input channels
patch_size (int): patch size
num_levels (int): number of block hierarchies (T_d in the paper)
embed_dims (int, tuple): embedding dimensions of each level
num_heads (int, tuple): number of attention heads for each level
depths (int, tuple): number of transformer layers for each level
num_classes (int): number of classes for classification head
mlp_ratio (int): ratio of mlp hidden dim to embedding dim for MLP of transformer layers
qkv_bias (bool): enable bias for qkv if True
drop_rate (float): dropout rate for MLP of transformer layers, MSA final projection layer, and classifier
attn_drop_rate (float): attention dropout rate
drop_path_rate (float): stochastic depth rate
norm_layer: (nn.Module): normalization layer for transformer layers
act_layer: (nn.Module): activation layer in MLP of transformer layers
pad_type: str: Type of padding to use '' for PyTorch symmetric, 'same' for TF SAME
weight_init: (str): weight init scheme
global_pool: (str): type of pooling operation to apply to final feature map
Notes:
- Default values follow NesT-B from the original Jax code.
- `embed_dims`, `num_heads`, `depths` should be ints or tuples with length `num_levels`.
- For those following the paper, Table A1 may have errors!
- https://github.com/google-research/nested-transformer/issues/2
"""
super().__init__()
for param_name in ['embed_dims', 'num_heads', 'depths']:
param_value = locals()[param_name]
if isinstance(param_value, collections.abc.Sequence):
assert len(param_value) == num_levels, f'Require `len({param_name}) == num_levels`'
embed_dims = to_ntuple(num_levels)(embed_dims)
num_heads = to_ntuple(num_levels)(num_heads)
depths = to_ntuple(num_levels)(depths)
self.num_classes = num_classes
self.num_features = embed_dims[-1]
self.feature_info = []
norm_layer = norm_layer or LayerNorm
act_layer = act_layer or nn.GELU
self.drop_rate = drop_rate
self.num_levels = num_levels
if isinstance(img_size, collections.abc.Sequence):
assert img_size[0] == img_size[1], 'Model only handles square inputs'
img_size = img_size[0]
assert img_size % patch_size == 0, '`patch_size` must divide `img_size` evenly'
self.patch_size = patch_size
# Number of blocks at each level
self.num_blocks = (4 ** torch.arange(num_levels)).flip(0).tolist()
assert (img_size // patch_size) % math.sqrt(self.num_blocks[0]) == 0, \
'First level blocks don\'t fit evenly. Check `img_size`, `patch_size`, and `num_levels`'
# Block edge size in units of patches
# Hint: (img_size // patch_size) gives number of patches along edge of image. sqrt(self.num_blocks[0]) is the
# number of blocks along edge of image
self.block_size = int((img_size // patch_size) // math.sqrt(self.num_blocks[0]))
# Patch embedding
self.patch_embed = PatchEmbed(
img_size=img_size,
patch_size=patch_size,
in_chans=in_chans,
embed_dim=embed_dims[0],
flatten=False,
)
self.num_patches = self.patch_embed.num_patches
self.seq_length = self.num_patches // self.num_blocks[0]
# Build up each hierarchical level
levels = []
dp_rates = [x.tolist() for x in torch.linspace(0, drop_path_rate, sum(depths)).split(depths)]
prev_dim = None
curr_stride = 4
for i in range(len(self.num_blocks)):
dim = embed_dims[i]
levels.append(NestLevel(
self.num_blocks[i],
self.block_size,
self.seq_length,
num_heads[i],
depths[i],
dim,
prev_dim,
mlp_ratio=mlp_ratio,
qkv_bias=qkv_bias,
proj_drop=proj_drop_rate,
attn_drop=attn_drop_rate,
drop_path=dp_rates[i],
norm_layer=norm_layer,
act_layer=act_layer,
pad_type=pad_type,
))
self.feature_info += [dict(num_chs=dim, reduction=curr_stride, module=f'levels.{i}')]
prev_dim = dim
curr_stride *= 2
self.levels = nn.Sequential(*levels)
# Final normalization layer
self.norm = norm_layer(embed_dims[-1])
# Classifier
global_pool, head = create_classifier(self.num_features, self.num_classes, pool_type=global_pool)
self.global_pool = global_pool
self.head_drop = nn.Dropout(drop_rate)
self.head = head
self.init_weights(weight_init)
@torch.jit.ignore
def init_weights(self, mode=''):
assert mode in ('nlhb', '')
head_bias = -math.log(self.num_classes) if 'nlhb' in mode else 0.
for level in self.levels:
trunc_normal_(level.pos_embed, std=.02, a=-2, b=2)
named_apply(partial(_init_nest_weights, head_bias=head_bias), self)
@torch.jit.ignore
def no_weight_decay(self):
return {f'level.{i}.pos_embed' for i in range(len(self.levels))}
@torch.jit.ignore
def group_matcher(self, coarse=False):
matcher = dict(
stem=r'^patch_embed', # stem and embed
blocks=[
(r'^levels\.(\d+)' if coarse else r'^levels\.(\d+)\.transformer_encoder\.(\d+)', None),
(r'^levels\.(\d+)\.(?:pool|pos_embed)', (0,)),
(r'^norm', (99999,))
]
)
return matcher
@torch.jit.ignore
def set_grad_checkpointing(self, enable=True):
for l in self.levels:
l.grad_checkpointing = enable
@torch.jit.ignore
def get_classifier(self):
return self.head
def reset_classifier(self, num_classes, global_pool='avg'):
self.num_classes = num_classes
self.global_pool, self.head = create_classifier(
self.num_features, self.num_classes, pool_type=global_pool)
def forward_features(self, x):
x = self.patch_embed(x)
x = self.levels(x)
# Layer norm done over channel dim only (to NHWC and back)
x = self.norm(x.permute(0, 2, 3, 1)).permute(0, 3, 1, 2)
return x
def forward_head(self, x, pre_logits: bool = False):
x = self.global_pool(x)
x = self.head_drop(x)
return x if pre_logits else self.head(x)
def forward(self, x):
x = self.forward_features(x)
x = self.forward_head(x)
return x
def _init_nest_weights(module: nn.Module, name: str = '', head_bias: float = 0.):
""" NesT weight initialization
Can replicate Jax implementation. Otherwise follows vision_transformer.py
"""
if isinstance(module, nn.Linear):
if name.startswith('head'):
trunc_normal_(module.weight, std=.02, a=-2, b=2)
nn.init.constant_(module.bias, head_bias)
else:
trunc_normal_(module.weight, std=.02, a=-2, b=2)
if module.bias is not None:
nn.init.zeros_(module.bias)
elif isinstance(module, nn.Conv2d):
trunc_normal_(module.weight, std=.02, a=-2, b=2)
if module.bias is not None:
nn.init.zeros_(module.bias)
def resize_pos_embed(posemb, posemb_new):
"""
Rescale the grid of position embeddings when loading from state_dict
Expected shape of position embeddings is (1, T, N, C), and considers only square images
"""
_logger.info('Resized position embedding: %s to %s', posemb.shape, posemb_new.shape)
seq_length_old = posemb.shape[2]
num_blocks_new, seq_length_new = posemb_new.shape[1:3]
size_new = int(math.sqrt(num_blocks_new*seq_length_new))
# First change to (1, C, H, W)
posemb = deblockify(posemb, int(math.sqrt(seq_length_old))).permute(0, 3, 1, 2)
posemb = F.interpolate(posemb, size=[size_new, size_new], mode='bicubic', align_corners=False)
# Now change to new (1, T, N, C)
posemb = blockify(posemb.permute(0, 2, 3, 1), int(math.sqrt(seq_length_new)))
return posemb
def checkpoint_filter_fn(state_dict, model):
""" resize positional embeddings of pretrained weights """
pos_embed_keys = [k for k in state_dict.keys() if k.startswith('pos_embed_')]
for k in pos_embed_keys:
if state_dict[k].shape != getattr(model, k).shape:
state_dict[k] = resize_pos_embed(state_dict[k], getattr(model, k))
return state_dict
def _create_nest(variant, pretrained=False, **kwargs):
model = build_model_with_cfg(
Nest,
variant,
pretrained,
feature_cfg=dict(out_indices=(0, 1, 2), flatten_sequential=True),
pretrained_filter_fn=checkpoint_filter_fn,
**kwargs,
)
return model
def _cfg(url='', **kwargs):
return {
'url': url,
'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': [14, 14],
'crop_pct': .875, 'interpolation': 'bicubic', 'fixed_input_size': True,
'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
'first_conv': 'patch_embed.proj', 'classifier': 'head',
**kwargs
}
default_cfgs = generate_default_cfgs({
'nest_base.untrained': _cfg(),
'nest_small.untrained': _cfg(),
'nest_tiny.untrained': _cfg(),
# (weights from official Google JAX impl, require 'SAME' padding)
'nest_base_jx.goog_in1k': _cfg(hf_hub_id='timm/'),
'nest_small_jx.goog_in1k': _cfg(hf_hub_id='timm/'),
'nest_tiny_jx.goog_in1k': _cfg(hf_hub_id='timm/'),
})
@register_model
def nest_base(pretrained=False, **kwargs) -> Nest:
""" Nest-B @ 224x224
"""
model_kwargs = dict(
embed_dims=(128, 256, 512), num_heads=(4, 8, 16), depths=(2, 2, 20), **kwargs)
model = _create_nest('nest_base', pretrained=pretrained, **model_kwargs)
return model
@register_model
def nest_small(pretrained=False, **kwargs) -> Nest:
""" Nest-S @ 224x224
"""
model_kwargs = dict(embed_dims=(96, 192, 384), num_heads=(3, 6, 12), depths=(2, 2, 20), **kwargs)
model = _create_nest('nest_small', pretrained=pretrained, **model_kwargs)
return model
@register_model
def nest_tiny(pretrained=False, **kwargs) -> Nest:
""" Nest-T @ 224x224
"""
model_kwargs = dict(embed_dims=(96, 192, 384), num_heads=(3, 6, 12), depths=(2, 2, 8), **kwargs)
model = _create_nest('nest_tiny', pretrained=pretrained, **model_kwargs)
return model
@register_model
def nest_base_jx(pretrained=False, **kwargs) -> Nest:
""" Nest-B @ 224x224
"""
kwargs.setdefault('pad_type', 'same')
model_kwargs = dict(
embed_dims=(128, 256, 512), num_heads=(4, 8, 16), depths=(2, 2, 20), **kwargs)
model = _create_nest('nest_base_jx', pretrained=pretrained, **model_kwargs)
return model
@register_model
def nest_small_jx(pretrained=False, **kwargs) -> Nest:
""" Nest-S @ 224x224
"""
kwargs.setdefault('pad_type', 'same')
model_kwargs = dict(embed_dims=(96, 192, 384), num_heads=(3, 6, 12), depths=(2, 2, 20), **kwargs)
model = _create_nest('nest_small_jx', pretrained=pretrained, **model_kwargs)
return model
@register_model
def nest_tiny_jx(pretrained=False, **kwargs) -> Nest:
""" Nest-T @ 224x224
"""
kwargs.setdefault('pad_type', 'same')
model_kwargs = dict(embed_dims=(96, 192, 384), num_heads=(3, 6, 12), depths=(2, 2, 8), **kwargs)
model = _create_nest('nest_tiny_jx', pretrained=pretrained, **model_kwargs)
return model
register_model_deprecations(__name__, {
'jx_nest_base': 'nest_base_jx',
'jx_nest_small': 'nest_small_jx',
'jx_nest_tiny': 'nest_tiny_jx',
}) | pytorch-image-models/timm/models/nest.py/0 | {
"file_path": "pytorch-image-models/timm/models/nest.py",
"repo_id": "pytorch-image-models",
"token_count": 10075
} | 187 |
""" Sequencer
Paper: `Sequencer: Deep LSTM for Image Classification` - https://arxiv.org/pdf/2205.01972.pdf
"""
# Copyright (c) 2022. Yuki Tatsunami
# Licensed under the Apache License, Version 2.0 (the "License");
import math
from functools import partial
from itertools import accumulate
from typing import Tuple
import torch
import torch.nn as nn
from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD, DEFAULT_CROP_PCT
from timm.layers import lecun_normal_, DropPath, Mlp, PatchEmbed, ClassifierHead
from ._builder import build_model_with_cfg
from ._manipulate import named_apply
from ._registry import register_model, generate_default_cfgs
__all__ = ['Sequencer2d'] # model_registry will add each entrypoint fn to this
def _init_weights(module: nn.Module, name: str, head_bias: float = 0., flax=False):
if isinstance(module, nn.Linear):
if name.startswith('head'):
nn.init.zeros_(module.weight)
nn.init.constant_(module.bias, head_bias)
else:
if flax:
# Flax defaults
lecun_normal_(module.weight)
if module.bias is not None:
nn.init.zeros_(module.bias)
else:
nn.init.xavier_uniform_(module.weight)
if module.bias is not None:
if 'mlp' in name:
nn.init.normal_(module.bias, std=1e-6)
else:
nn.init.zeros_(module.bias)
elif isinstance(module, nn.Conv2d):
lecun_normal_(module.weight)
if module.bias is not None:
nn.init.zeros_(module.bias)
elif isinstance(module, (nn.LayerNorm, nn.BatchNorm2d, nn.GroupNorm)):
nn.init.ones_(module.weight)
nn.init.zeros_(module.bias)
elif isinstance(module, (nn.RNN, nn.GRU, nn.LSTM)):
stdv = 1.0 / math.sqrt(module.hidden_size)
for weight in module.parameters():
nn.init.uniform_(weight, -stdv, stdv)
elif hasattr(module, 'init_weights'):
module.init_weights()
class RNNIdentity(nn.Module):
def __init__(self, *args, **kwargs):
super(RNNIdentity, self).__init__()
def forward(self, x: torch.Tensor) -> Tuple[torch.Tensor, None]:
return x, None
class RNN2dBase(nn.Module):
def __init__(
self,
input_size: int,
hidden_size: int,
num_layers: int = 1,
bias: bool = True,
bidirectional: bool = True,
union="cat",
with_fc=True,
):
super().__init__()
self.input_size = input_size
self.hidden_size = hidden_size
self.output_size = 2 * hidden_size if bidirectional else hidden_size
self.union = union
self.with_vertical = True
self.with_horizontal = True
self.with_fc = with_fc
self.fc = None
if with_fc:
if union == "cat":
self.fc = nn.Linear(2 * self.output_size, input_size)
elif union == "add":
self.fc = nn.Linear(self.output_size, input_size)
elif union == "vertical":
self.fc = nn.Linear(self.output_size, input_size)
self.with_horizontal = False
elif union == "horizontal":
self.fc = nn.Linear(self.output_size, input_size)
self.with_vertical = False
else:
raise ValueError("Unrecognized union: " + union)
elif union == "cat":
pass
if 2 * self.output_size != input_size:
raise ValueError(f"The output channel {2 * self.output_size} is different from the input channel {input_size}.")
elif union == "add":
pass
if self.output_size != input_size:
raise ValueError(f"The output channel {self.output_size} is different from the input channel {input_size}.")
elif union == "vertical":
if self.output_size != input_size:
raise ValueError(f"The output channel {self.output_size} is different from the input channel {input_size}.")
self.with_horizontal = False
elif union == "horizontal":
if self.output_size != input_size:
raise ValueError(f"The output channel {self.output_size} is different from the input channel {input_size}.")
self.with_vertical = False
else:
raise ValueError("Unrecognized union: " + union)
self.rnn_v = RNNIdentity()
self.rnn_h = RNNIdentity()
def forward(self, x):
B, H, W, C = x.shape
if self.with_vertical:
v = x.permute(0, 2, 1, 3)
v = v.reshape(-1, H, C)
v, _ = self.rnn_v(v)
v = v.reshape(B, W, H, -1)
v = v.permute(0, 2, 1, 3)
else:
v = None
if self.with_horizontal:
h = x.reshape(-1, W, C)
h, _ = self.rnn_h(h)
h = h.reshape(B, H, W, -1)
else:
h = None
if v is not None and h is not None:
if self.union == "cat":
x = torch.cat([v, h], dim=-1)
else:
x = v + h
elif v is not None:
x = v
elif h is not None:
x = h
if self.fc is not None:
x = self.fc(x)
return x
class LSTM2d(RNN2dBase):
def __init__(
self,
input_size: int,
hidden_size: int,
num_layers: int = 1,
bias: bool = True,
bidirectional: bool = True,
union="cat",
with_fc=True,
):
super().__init__(input_size, hidden_size, num_layers, bias, bidirectional, union, with_fc)
if self.with_vertical:
self.rnn_v = nn.LSTM(
input_size,
hidden_size,
num_layers,
batch_first=True,
bias=bias,
bidirectional=bidirectional,
)
if self.with_horizontal:
self.rnn_h = nn.LSTM(
input_size,
hidden_size,
num_layers,
batch_first=True,
bias=bias,
bidirectional=bidirectional,
)
class Sequencer2dBlock(nn.Module):
def __init__(
self,
dim,
hidden_size,
mlp_ratio=3.0,
rnn_layer=LSTM2d,
mlp_layer=Mlp,
norm_layer=partial(nn.LayerNorm, eps=1e-6),
act_layer=nn.GELU,
num_layers=1,
bidirectional=True,
union="cat",
with_fc=True,
drop=0.,
drop_path=0.,
):
super().__init__()
channels_dim = int(mlp_ratio * dim)
self.norm1 = norm_layer(dim)
self.rnn_tokens = rnn_layer(
dim,
hidden_size,
num_layers=num_layers,
bidirectional=bidirectional,
union=union,
with_fc=with_fc,
)
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
self.norm2 = norm_layer(dim)
self.mlp_channels = mlp_layer(dim, channels_dim, act_layer=act_layer, drop=drop)
def forward(self, x):
x = x + self.drop_path(self.rnn_tokens(self.norm1(x)))
x = x + self.drop_path(self.mlp_channels(self.norm2(x)))
return x
class Shuffle(nn.Module):
def __init__(self):
super().__init__()
def forward(self, x):
if self.training:
B, H, W, C = x.shape
r = torch.randperm(H * W)
x = x.reshape(B, -1, C)
x = x[:, r, :].reshape(B, H, W, -1)
return x
class Downsample2d(nn.Module):
def __init__(self, input_dim, output_dim, patch_size):
super().__init__()
self.down = nn.Conv2d(input_dim, output_dim, kernel_size=patch_size, stride=patch_size)
def forward(self, x):
x = x.permute(0, 3, 1, 2)
x = self.down(x)
x = x.permute(0, 2, 3, 1)
return x
class Sequencer2dStage(nn.Module):
def __init__(
self,
dim,
dim_out,
depth,
patch_size,
hidden_size,
mlp_ratio,
downsample=False,
block_layer=Sequencer2dBlock,
rnn_layer=LSTM2d,
mlp_layer=Mlp,
norm_layer=partial(nn.LayerNorm, eps=1e-6),
act_layer=nn.GELU,
num_layers=1,
bidirectional=True,
union="cat",
with_fc=True,
drop=0.,
drop_path=0.,
):
super().__init__()
if downsample:
self.downsample = Downsample2d(dim, dim_out, patch_size)
else:
assert dim == dim_out
self.downsample = nn.Identity()
blocks = []
for block_idx in range(depth):
blocks.append(block_layer(
dim_out,
hidden_size,
mlp_ratio=mlp_ratio,
rnn_layer=rnn_layer,
mlp_layer=mlp_layer,
norm_layer=norm_layer,
act_layer=act_layer,
num_layers=num_layers,
bidirectional=bidirectional,
union=union,
with_fc=with_fc,
drop=drop,
drop_path=drop_path[block_idx] if isinstance(drop_path, (list, tuple)) else drop_path,
))
self.blocks = nn.Sequential(*blocks)
def forward(self, x):
x = self.downsample(x)
x = self.blocks(x)
return x
class Sequencer2d(nn.Module):
def __init__(
self,
num_classes=1000,
img_size=224,
in_chans=3,
global_pool='avg',
layers=(4, 3, 8, 3),
patch_sizes=(7, 2, 2, 1),
embed_dims=(192, 384, 384, 384),
hidden_sizes=(48, 96, 96, 96),
mlp_ratios=(3.0, 3.0, 3.0, 3.0),
block_layer=Sequencer2dBlock,
rnn_layer=LSTM2d,
mlp_layer=Mlp,
norm_layer=partial(nn.LayerNorm, eps=1e-6),
act_layer=nn.GELU,
num_rnn_layers=1,
bidirectional=True,
union="cat",
with_fc=True,
drop_rate=0.,
drop_path_rate=0.,
nlhb=False,
stem_norm=False,
):
super().__init__()
assert global_pool in ('', 'avg')
self.num_classes = num_classes
self.global_pool = global_pool
self.num_features = embed_dims[-1] # num_features for consistency with other models
self.feature_dim = -1 # channel dim index for feature outputs (rank 4, NHWC)
self.output_fmt = 'NHWC'
self.feature_info = []
self.stem = PatchEmbed(
img_size=None,
patch_size=patch_sizes[0],
in_chans=in_chans,
embed_dim=embed_dims[0],
norm_layer=norm_layer if stem_norm else None,
flatten=False,
output_fmt='NHWC',
)
assert len(layers) == len(patch_sizes) == len(embed_dims) == len(hidden_sizes) == len(mlp_ratios)
reductions = list(accumulate(patch_sizes, lambda x, y: x * y))
stages = []
prev_dim = embed_dims[0]
for i, _ in enumerate(embed_dims):
stages += [Sequencer2dStage(
prev_dim,
embed_dims[i],
depth=layers[i],
downsample=i > 0,
patch_size=patch_sizes[i],
hidden_size=hidden_sizes[i],
mlp_ratio=mlp_ratios[i],
block_layer=block_layer,
rnn_layer=rnn_layer,
mlp_layer=mlp_layer,
norm_layer=norm_layer,
act_layer=act_layer,
num_layers=num_rnn_layers,
bidirectional=bidirectional,
union=union,
with_fc=with_fc,
drop=drop_rate,
drop_path=drop_path_rate,
)]
prev_dim = embed_dims[i]
self.feature_info += [dict(num_chs=prev_dim, reduction=reductions[i], module=f'stages.{i}')]
self.stages = nn.Sequential(*stages)
self.norm = norm_layer(embed_dims[-1])
self.head = ClassifierHead(
self.num_features,
num_classes,
pool_type=global_pool,
drop_rate=drop_rate,
input_fmt=self.output_fmt,
)
self.init_weights(nlhb=nlhb)
def init_weights(self, nlhb=False):
head_bias = -math.log(self.num_classes) if nlhb else 0.
named_apply(partial(_init_weights, head_bias=head_bias), module=self) # depth-first
@torch.jit.ignore
def group_matcher(self, coarse=False):
return dict(
stem=r'^stem',
blocks=[
(r'^stages\.(\d+)', None),
(r'^norm', (99999,))
] if coarse else [
(r'^stages\.(\d+)\.blocks\.(\d+)', None),
(r'^stages\.(\d+)\.downsample', (0,)),
(r'^norm', (99999,))
]
)
@torch.jit.ignore
def set_grad_checkpointing(self, enable=True):
assert not enable, 'gradient checkpointing not supported'
@torch.jit.ignore
def get_classifier(self):
return self.head
def reset_classifier(self, num_classes, global_pool=None):
self.num_classes = num_classes
self.head.reset(num_classes, pool_type=global_pool)
def forward_features(self, x):
x = self.stem(x)
x = self.stages(x)
x = self.norm(x)
return x
def forward_head(self, x, pre_logits: bool = False):
return self.head(x, pre_logits=True) if pre_logits else self.head(x)
def forward(self, x):
x = self.forward_features(x)
x = self.forward_head(x)
return x
def checkpoint_filter_fn(state_dict, model):
""" Remap original checkpoints -> timm """
if 'stages.0.blocks.0.norm1.weight' in state_dict:
return state_dict # already translated checkpoint
if 'model' in state_dict:
state_dict = state_dict['model']
import re
out_dict = {}
for k, v in state_dict.items():
k = re.sub(r'blocks.([0-9]+).([0-9]+).down', lambda x: f'stages.{int(x.group(1)) + 1}.downsample.down', k)
k = re.sub(r'blocks.([0-9]+).([0-9]+)', r'stages.\1.blocks.\2', k)
k = k.replace('head.', 'head.fc.')
out_dict[k] = v
return out_dict
def _create_sequencer2d(variant, pretrained=False, **kwargs):
default_out_indices = tuple(range(3))
out_indices = kwargs.pop('out_indices', default_out_indices)
model = build_model_with_cfg(
Sequencer2d,
variant,
pretrained,
pretrained_filter_fn=checkpoint_filter_fn,
feature_cfg=dict(flatten_sequential=True, out_indices=out_indices),
**kwargs,
)
return model
def _cfg(url='', **kwargs):
return {
'url': url,
'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None,
'crop_pct': DEFAULT_CROP_PCT, 'interpolation': 'bicubic', 'fixed_input_size': True,
'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
'first_conv': 'stem.proj', 'classifier': 'head.fc',
**kwargs
}
default_cfgs = generate_default_cfgs({
'sequencer2d_s.in1k': _cfg(hf_hub_id='timm/'),
'sequencer2d_m.in1k': _cfg(hf_hub_id='timm/'),
'sequencer2d_l.in1k': _cfg(hf_hub_id='timm/'),
})
@register_model
def sequencer2d_s(pretrained=False, **kwargs) -> Sequencer2d:
model_args = dict(
layers=[4, 3, 8, 3],
patch_sizes=[7, 2, 1, 1],
embed_dims=[192, 384, 384, 384],
hidden_sizes=[48, 96, 96, 96],
mlp_ratios=[3.0, 3.0, 3.0, 3.0],
rnn_layer=LSTM2d,
bidirectional=True,
union="cat",
with_fc=True,
)
model = _create_sequencer2d('sequencer2d_s', pretrained=pretrained, **dict(model_args, **kwargs))
return model
@register_model
def sequencer2d_m(pretrained=False, **kwargs) -> Sequencer2d:
model_args = dict(
layers=[4, 3, 14, 3],
patch_sizes=[7, 2, 1, 1],
embed_dims=[192, 384, 384, 384],
hidden_sizes=[48, 96, 96, 96],
mlp_ratios=[3.0, 3.0, 3.0, 3.0],
rnn_layer=LSTM2d,
bidirectional=True,
union="cat",
with_fc=True,
**kwargs)
model = _create_sequencer2d('sequencer2d_m', pretrained=pretrained, **dict(model_args, **kwargs))
return model
@register_model
def sequencer2d_l(pretrained=False, **kwargs) -> Sequencer2d:
model_args = dict(
layers=[8, 8, 16, 4],
patch_sizes=[7, 2, 1, 1],
embed_dims=[192, 384, 384, 384],
hidden_sizes=[48, 96, 96, 96],
mlp_ratios=[3.0, 3.0, 3.0, 3.0],
rnn_layer=LSTM2d,
bidirectional=True,
union="cat",
with_fc=True,
**kwargs)
model = _create_sequencer2d('sequencer2d_l', pretrained=pretrained, **dict(model_args, **kwargs))
return model
| pytorch-image-models/timm/models/sequencer.py/0 | {
"file_path": "pytorch-image-models/timm/models/sequencer.py",
"repo_id": "pytorch-image-models",
"token_count": 9227
} | 188 |
""" VoVNet (V1 & V2)
Papers:
* `An Energy and GPU-Computation Efficient Backbone Network` - https://arxiv.org/abs/1904.09730
* `CenterMask : Real-Time Anchor-Free Instance Segmentation` - https://arxiv.org/abs/1911.06667
Looked at https://github.com/youngwanLEE/vovnet-detectron2 &
https://github.com/stigma0617/VoVNet.pytorch/blob/master/models_vovnet/vovnet.py
for some reference, rewrote most of the code.
Hacked together by / Copyright 2020 Ross Wightman
"""
from typing import List
import torch
import torch.nn as nn
from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
from timm.layers import ConvNormAct, SeparableConvNormAct, BatchNormAct2d, ClassifierHead, DropPath, \
create_attn, create_norm_act_layer
from ._builder import build_model_with_cfg
from ._manipulate import checkpoint_seq
from ._registry import register_model, generate_default_cfgs
__all__ = ['VovNet'] # model_registry will add each entrypoint fn to this
class SequentialAppendList(nn.Sequential):
def __init__(self, *args):
super(SequentialAppendList, self).__init__(*args)
def forward(self, x: torch.Tensor, concat_list: List[torch.Tensor]) -> torch.Tensor:
for i, module in enumerate(self):
if i == 0:
concat_list.append(module(x))
else:
concat_list.append(module(concat_list[-1]))
x = torch.cat(concat_list, dim=1)
return x
class OsaBlock(nn.Module):
def __init__(
self,
in_chs,
mid_chs,
out_chs,
layer_per_block,
residual=False,
depthwise=False,
attn='',
norm_layer=BatchNormAct2d,
act_layer=nn.ReLU,
drop_path=None,
):
super(OsaBlock, self).__init__()
self.residual = residual
self.depthwise = depthwise
conv_kwargs = dict(norm_layer=norm_layer, act_layer=act_layer)
next_in_chs = in_chs
if self.depthwise and next_in_chs != mid_chs:
assert not residual
self.conv_reduction = ConvNormAct(next_in_chs, mid_chs, 1, **conv_kwargs)
else:
self.conv_reduction = None
mid_convs = []
for i in range(layer_per_block):
if self.depthwise:
conv = SeparableConvNormAct(mid_chs, mid_chs, **conv_kwargs)
else:
conv = ConvNormAct(next_in_chs, mid_chs, 3, **conv_kwargs)
next_in_chs = mid_chs
mid_convs.append(conv)
self.conv_mid = SequentialAppendList(*mid_convs)
# feature aggregation
next_in_chs = in_chs + layer_per_block * mid_chs
self.conv_concat = ConvNormAct(next_in_chs, out_chs, **conv_kwargs)
self.attn = create_attn(attn, out_chs) if attn else None
self.drop_path = drop_path
def forward(self, x):
output = [x]
if self.conv_reduction is not None:
x = self.conv_reduction(x)
x = self.conv_mid(x, output)
x = self.conv_concat(x)
if self.attn is not None:
x = self.attn(x)
if self.drop_path is not None:
x = self.drop_path(x)
if self.residual:
x = x + output[0]
return x
class OsaStage(nn.Module):
def __init__(
self,
in_chs,
mid_chs,
out_chs,
block_per_stage,
layer_per_block,
downsample=True,
residual=True,
depthwise=False,
attn='ese',
norm_layer=BatchNormAct2d,
act_layer=nn.ReLU,
drop_path_rates=None,
):
super(OsaStage, self).__init__()
self.grad_checkpointing = False
if downsample:
self.pool = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
else:
self.pool = None
blocks = []
for i in range(block_per_stage):
last_block = i == block_per_stage - 1
if drop_path_rates is not None and drop_path_rates[i] > 0.:
drop_path = DropPath(drop_path_rates[i])
else:
drop_path = None
blocks += [OsaBlock(
in_chs, mid_chs, out_chs, layer_per_block, residual=residual and i > 0, depthwise=depthwise,
attn=attn if last_block else '', norm_layer=norm_layer, act_layer=act_layer, drop_path=drop_path)
]
in_chs = out_chs
self.blocks = nn.Sequential(*blocks)
def forward(self, x):
if self.pool is not None:
x = self.pool(x)
if self.grad_checkpointing and not torch.jit.is_scripting():
x = checkpoint_seq(self.blocks, x)
else:
x = self.blocks(x)
return x
class VovNet(nn.Module):
def __init__(
self,
cfg,
in_chans=3,
num_classes=1000,
global_pool='avg',
output_stride=32,
norm_layer=BatchNormAct2d,
act_layer=nn.ReLU,
drop_rate=0.,
drop_path_rate=0.,
**kwargs,
):
"""
Args:
cfg (dict): Model architecture configuration
in_chans (int): Number of input channels (default: 3)
num_classes (int): Number of classifier classes (default: 1000)
global_pool (str): Global pooling type (default: 'avg')
output_stride (int): Output stride of network, one of (8, 16, 32) (default: 32)
norm_layer (Union[str, nn.Module]): normalization layer
act_layer (Union[str, nn.Module]): activation layer
drop_rate (float): Dropout rate (default: 0.)
drop_path_rate (float): Stochastic depth drop-path rate (default: 0.)
kwargs (dict): Extra kwargs overlayed onto cfg
"""
super(VovNet, self).__init__()
self.num_classes = num_classes
self.drop_rate = drop_rate
assert output_stride == 32 # FIXME support dilation
cfg = dict(cfg, **kwargs)
stem_stride = cfg.get("stem_stride", 4)
stem_chs = cfg["stem_chs"]
stage_conv_chs = cfg["stage_conv_chs"]
stage_out_chs = cfg["stage_out_chs"]
block_per_stage = cfg["block_per_stage"]
layer_per_block = cfg["layer_per_block"]
conv_kwargs = dict(norm_layer=norm_layer, act_layer=act_layer)
# Stem module
last_stem_stride = stem_stride // 2
conv_type = SeparableConvNormAct if cfg["depthwise"] else ConvNormAct
self.stem = nn.Sequential(*[
ConvNormAct(in_chans, stem_chs[0], 3, stride=2, **conv_kwargs),
conv_type(stem_chs[0], stem_chs[1], 3, stride=1, **conv_kwargs),
conv_type(stem_chs[1], stem_chs[2], 3, stride=last_stem_stride, **conv_kwargs),
])
self.feature_info = [dict(
num_chs=stem_chs[1], reduction=2, module=f'stem.{1 if stem_stride == 4 else 2}')]
current_stride = stem_stride
# OSA stages
stage_dpr = torch.split(torch.linspace(0, drop_path_rate, sum(block_per_stage)), block_per_stage)
in_ch_list = stem_chs[-1:] + stage_out_chs[:-1]
stage_args = dict(residual=cfg["residual"], depthwise=cfg["depthwise"], attn=cfg["attn"], **conv_kwargs)
stages = []
for i in range(4): # num_stages
downsample = stem_stride == 2 or i > 0 # first stage has no stride/downsample if stem_stride is 4
stages += [OsaStage(
in_ch_list[i],
stage_conv_chs[i],
stage_out_chs[i],
block_per_stage[i],
layer_per_block,
downsample=downsample,
drop_path_rates=stage_dpr[i],
**stage_args,
)]
self.num_features = stage_out_chs[i]
current_stride *= 2 if downsample else 1
self.feature_info += [dict(num_chs=self.num_features, reduction=current_stride, module=f'stages.{i}')]
self.stages = nn.Sequential(*stages)
self.head = ClassifierHead(self.num_features, num_classes, pool_type=global_pool, drop_rate=drop_rate)
for n, m in self.named_modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
elif isinstance(m, nn.Linear):
nn.init.zeros_(m.bias)
@torch.jit.ignore
def group_matcher(self, coarse=False):
return dict(
stem=r'^stem',
blocks=r'^stages\.(\d+)' if coarse else r'^stages\.(\d+).blocks\.(\d+)',
)
@torch.jit.ignore
def set_grad_checkpointing(self, enable=True):
for s in self.stages:
s.grad_checkpointing = enable
@torch.jit.ignore
def get_classifier(self):
return self.head.fc
def reset_classifier(self, num_classes, global_pool='avg'):
self.head = ClassifierHead(self.num_features, num_classes, pool_type=global_pool, drop_rate=self.drop_rate)
def forward_features(self, x):
x = self.stem(x)
return self.stages(x)
def forward_head(self, x, pre_logits: bool = False):
return self.head(x, pre_logits=pre_logits)
def forward(self, x):
x = self.forward_features(x)
x = self.forward_head(x)
return x
# model cfgs adapted from https://github.com/youngwanLEE/vovnet-detectron2 &
# https://github.com/stigma0617/VoVNet.pytorch/blob/master/models_vovnet/vovnet.py
model_cfgs = dict(
vovnet39a=dict(
stem_chs=[64, 64, 128],
stage_conv_chs=[128, 160, 192, 224],
stage_out_chs=[256, 512, 768, 1024],
layer_per_block=5,
block_per_stage=[1, 1, 2, 2],
residual=False,
depthwise=False,
attn='',
),
vovnet57a=dict(
stem_chs=[64, 64, 128],
stage_conv_chs=[128, 160, 192, 224],
stage_out_chs=[256, 512, 768, 1024],
layer_per_block=5,
block_per_stage=[1, 1, 4, 3],
residual=False,
depthwise=False,
attn='',
),
ese_vovnet19b_slim_dw=dict(
stem_chs=[64, 64, 64],
stage_conv_chs=[64, 80, 96, 112],
stage_out_chs=[112, 256, 384, 512],
layer_per_block=3,
block_per_stage=[1, 1, 1, 1],
residual=True,
depthwise=True,
attn='ese',
),
ese_vovnet19b_dw=dict(
stem_chs=[64, 64, 64],
stage_conv_chs=[128, 160, 192, 224],
stage_out_chs=[256, 512, 768, 1024],
layer_per_block=3,
block_per_stage=[1, 1, 1, 1],
residual=True,
depthwise=True,
attn='ese',
),
ese_vovnet19b_slim=dict(
stem_chs=[64, 64, 128],
stage_conv_chs=[64, 80, 96, 112],
stage_out_chs=[112, 256, 384, 512],
layer_per_block=3,
block_per_stage=[1, 1, 1, 1],
residual=True,
depthwise=False,
attn='ese',
),
ese_vovnet19b=dict(
stem_chs=[64, 64, 128],
stage_conv_chs=[128, 160, 192, 224],
stage_out_chs=[256, 512, 768, 1024],
layer_per_block=3,
block_per_stage=[1, 1, 1, 1],
residual=True,
depthwise=False,
attn='ese',
),
ese_vovnet39b=dict(
stem_chs=[64, 64, 128],
stage_conv_chs=[128, 160, 192, 224],
stage_out_chs=[256, 512, 768, 1024],
layer_per_block=5,
block_per_stage=[1, 1, 2, 2],
residual=True,
depthwise=False,
attn='ese',
),
ese_vovnet57b=dict(
stem_chs=[64, 64, 128],
stage_conv_chs=[128, 160, 192, 224],
stage_out_chs=[256, 512, 768, 1024],
layer_per_block=5,
block_per_stage=[1, 1, 4, 3],
residual=True,
depthwise=False,
attn='ese',
),
ese_vovnet99b=dict(
stem_chs=[64, 64, 128],
stage_conv_chs=[128, 160, 192, 224],
stage_out_chs=[256, 512, 768, 1024],
layer_per_block=5,
block_per_stage=[1, 3, 9, 3],
residual=True,
depthwise=False,
attn='ese',
),
eca_vovnet39b=dict(
stem_chs=[64, 64, 128],
stage_conv_chs=[128, 160, 192, 224],
stage_out_chs=[256, 512, 768, 1024],
layer_per_block=5,
block_per_stage=[1, 1, 2, 2],
residual=True,
depthwise=False,
attn='eca',
),
)
model_cfgs['ese_vovnet39b_evos'] = model_cfgs['ese_vovnet39b']
def _create_vovnet(variant, pretrained=False, **kwargs):
return build_model_with_cfg(
VovNet,
variant,
pretrained,
model_cfg=model_cfgs[variant],
feature_cfg=dict(flatten_sequential=True),
**kwargs,
)
def _cfg(url='', **kwargs):
return {
'url': url, 'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': (7, 7),
'crop_pct': 0.875, 'interpolation': 'bicubic',
'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
'first_conv': 'stem.0.conv', 'classifier': 'head.fc', **kwargs,
}
default_cfgs = generate_default_cfgs({
'vovnet39a.untrained': _cfg(url=''),
'vovnet57a.untrained': _cfg(url=''),
'ese_vovnet19b_slim_dw.untrained': _cfg(url=''),
'ese_vovnet19b_dw.ra_in1k': _cfg(
hf_hub_id='timm/',
test_input_size=(3, 288, 288), test_crop_pct=0.95),
'ese_vovnet19b_slim.untrained': _cfg(url=''),
'ese_vovnet39b.ra_in1k': _cfg(
hf_hub_id='timm/',
test_input_size=(3, 288, 288), test_crop_pct=0.95),
'ese_vovnet57b.untrained': _cfg(url=''),
'ese_vovnet99b.untrained': _cfg(url=''),
'eca_vovnet39b.untrained': _cfg(url=''),
'ese_vovnet39b_evos.untrained': _cfg(url=''),
})
@register_model
def vovnet39a(pretrained=False, **kwargs) -> VovNet:
return _create_vovnet('vovnet39a', pretrained=pretrained, **kwargs)
@register_model
def vovnet57a(pretrained=False, **kwargs) -> VovNet:
return _create_vovnet('vovnet57a', pretrained=pretrained, **kwargs)
@register_model
def ese_vovnet19b_slim_dw(pretrained=False, **kwargs) -> VovNet:
return _create_vovnet('ese_vovnet19b_slim_dw', pretrained=pretrained, **kwargs)
@register_model
def ese_vovnet19b_dw(pretrained=False, **kwargs) -> VovNet:
return _create_vovnet('ese_vovnet19b_dw', pretrained=pretrained, **kwargs)
@register_model
def ese_vovnet19b_slim(pretrained=False, **kwargs) -> VovNet:
return _create_vovnet('ese_vovnet19b_slim', pretrained=pretrained, **kwargs)
@register_model
def ese_vovnet39b(pretrained=False, **kwargs) -> VovNet:
return _create_vovnet('ese_vovnet39b', pretrained=pretrained, **kwargs)
@register_model
def ese_vovnet57b(pretrained=False, **kwargs) -> VovNet:
return _create_vovnet('ese_vovnet57b', pretrained=pretrained, **kwargs)
@register_model
def ese_vovnet99b(pretrained=False, **kwargs) -> VovNet:
return _create_vovnet('ese_vovnet99b', pretrained=pretrained, **kwargs)
@register_model
def eca_vovnet39b(pretrained=False, **kwargs) -> VovNet:
return _create_vovnet('eca_vovnet39b', pretrained=pretrained, **kwargs)
# Experimental Models
@register_model
def ese_vovnet39b_evos(pretrained=False, **kwargs) -> VovNet:
def norm_act_fn(num_features, **nkwargs):
return create_norm_act_layer('evonorms0', num_features, jit=False, **nkwargs)
return _create_vovnet('ese_vovnet39b_evos', pretrained=pretrained, norm_layer=norm_act_fn, **kwargs)
| pytorch-image-models/timm/models/vovnet.py/0 | {
"file_path": "pytorch-image-models/timm/models/vovnet.py",
"repo_id": "pytorch-image-models",
"token_count": 7769
} | 189 |
import math
import torch
from torch.optim.optimizer import Optimizer
class Nadam(Optimizer):
"""Implements Nadam algorithm (a variant of Adam based on Nesterov momentum).
It has been proposed in `Incorporating Nesterov Momentum into Adam`__.
Arguments:
params (iterable): iterable of parameters to optimize or dicts defining
parameter groups
lr (float, optional): learning rate (default: 2e-3)
betas (Tuple[float, float], optional): coefficients used for computing
running averages of gradient and its square
eps (float, optional): term added to the denominator to improve
numerical stability (default: 1e-8)
weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
schedule_decay (float, optional): momentum schedule decay (default: 4e-3)
__ http://cs229.stanford.edu/proj2015/054_report.pdf
__ http://www.cs.toronto.edu/~fritz/absps/momentum.pdf
Originally taken from: https://github.com/pytorch/pytorch/pull/1408
NOTE: Has potential issues but does work well on some problems.
"""
def __init__(self, params, lr=2e-3, betas=(0.9, 0.999), eps=1e-8,
weight_decay=0, schedule_decay=4e-3):
if not 0.0 <= lr:
raise ValueError("Invalid learning rate: {}".format(lr))
defaults = dict(
lr=lr,
betas=betas,
eps=eps,
weight_decay=weight_decay,
schedule_decay=schedule_decay,
)
super(Nadam, self).__init__(params, defaults)
@torch.no_grad()
def step(self, closure=None):
"""Performs a single optimization step.
Arguments:
closure (callable, optional): A closure that reevaluates the model
and returns the loss.
"""
loss = None
if closure is not None:
with torch.enable_grad():
loss = closure()
for group in self.param_groups:
for p in group['params']:
if p.grad is None:
continue
grad = p.grad
state = self.state[p]
# State initialization
if len(state) == 0:
state['step'] = 0
state['m_schedule'] = 1.
state['exp_avg'] = torch.zeros_like(p)
state['exp_avg_sq'] = torch.zeros_like(p)
# Warming momentum schedule
m_schedule = state['m_schedule']
schedule_decay = group['schedule_decay']
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
beta1, beta2 = group['betas']
eps = group['eps']
state['step'] += 1
t = state['step']
bias_correction2 = 1 - beta2 ** t
if group['weight_decay'] != 0:
grad = grad.add(p, alpha=group['weight_decay'])
momentum_cache_t = beta1 * (1. - 0.5 * (0.96 ** (t * schedule_decay)))
momentum_cache_t_1 = beta1 * (1. - 0.5 * (0.96 ** ((t + 1) * schedule_decay)))
m_schedule_new = m_schedule * momentum_cache_t
m_schedule_next = m_schedule * momentum_cache_t * momentum_cache_t_1
state['m_schedule'] = m_schedule_new
# Decay the first and second moment running average coefficient
exp_avg.mul_(beta1).add_(grad, alpha=1. - beta1)
exp_avg_sq.mul_(beta2).addcmul_(grad, grad, value=1. - beta2)
denom = (exp_avg_sq.sqrt() / math.sqrt(bias_correction2)).add_(eps)
p.addcdiv_(grad, denom, value=-group['lr'] * (1. - momentum_cache_t) / (1. - m_schedule_new))
p.addcdiv_(exp_avg, denom, value=-group['lr'] * momentum_cache_t_1 / (1. - m_schedule_next))
return loss
| pytorch-image-models/timm/optim/nadam.py/0 | {
"file_path": "pytorch-image-models/timm/optim/nadam.py",
"repo_id": "pytorch-image-models",
"token_count": 1921
} | 190 |
""" TanH Scheduler
TanH schedule with warmup, cycle/restarts, noise.
Hacked together by / Copyright 2021 Ross Wightman
"""
import logging
import math
import numpy as np
import torch
from .scheduler import Scheduler
_logger = logging.getLogger(__name__)
class TanhLRScheduler(Scheduler):
"""
Hyberbolic-Tangent decay with restarts.
This is described in the paper https://arxiv.org/abs/1806.01593
"""
def __init__(
self,
optimizer: torch.optim.Optimizer,
t_initial: int,
lb: float = -7.,
ub: float = 3.,
lr_min: float = 0.,
cycle_mul: float = 1.,
cycle_decay: float = 1.,
cycle_limit: int = 1,
warmup_t=0,
warmup_lr_init=0,
warmup_prefix=False,
t_in_epochs=True,
noise_range_t=None,
noise_pct=0.67,
noise_std=1.0,
noise_seed=42,
initialize=True,
) -> None:
super().__init__(
optimizer,
param_group_field="lr",
t_in_epochs=t_in_epochs,
noise_range_t=noise_range_t,
noise_pct=noise_pct,
noise_std=noise_std,
noise_seed=noise_seed,
initialize=initialize,
)
assert t_initial > 0
assert lr_min >= 0
assert lb < ub
assert cycle_limit >= 0
assert warmup_t >= 0
assert warmup_lr_init >= 0
self.lb = lb
self.ub = ub
self.t_initial = t_initial
self.lr_min = lr_min
self.cycle_mul = cycle_mul
self.cycle_decay = cycle_decay
self.cycle_limit = cycle_limit
self.warmup_t = warmup_t
self.warmup_lr_init = warmup_lr_init
self.warmup_prefix = warmup_prefix
if self.warmup_t:
t_v = self.base_values if self.warmup_prefix else self._get_lr(self.warmup_t)
self.warmup_steps = [(v - warmup_lr_init) / self.warmup_t for v in t_v]
super().update_groups(self.warmup_lr_init)
else:
self.warmup_steps = [1 for _ in self.base_values]
def _get_lr(self, t):
if t < self.warmup_t:
lrs = [self.warmup_lr_init + t * s for s in self.warmup_steps]
else:
if self.warmup_prefix:
t = t - self.warmup_t
if self.cycle_mul != 1:
i = math.floor(math.log(1 - t / self.t_initial * (1 - self.cycle_mul), self.cycle_mul))
t_i = self.cycle_mul ** i * self.t_initial
t_curr = t - (1 - self.cycle_mul ** i) / (1 - self.cycle_mul) * self.t_initial
else:
i = t // self.t_initial
t_i = self.t_initial
t_curr = t - (self.t_initial * i)
if i < self.cycle_limit:
gamma = self.cycle_decay ** i
lr_max_values = [v * gamma for v in self.base_values]
tr = t_curr / t_i
lrs = [
self.lr_min + 0.5 * (lr_max - self.lr_min) * (1 - math.tanh(self.lb * (1. - tr) + self.ub * tr))
for lr_max in lr_max_values
]
else:
lrs = [self.lr_min for _ in self.base_values]
return lrs
def get_cycle_length(self, cycles=0):
cycles = max(1, cycles or self.cycle_limit)
if self.cycle_mul == 1.0:
return self.t_initial * cycles
else:
return int(math.floor(-self.t_initial * (self.cycle_mul ** cycles - 1) / (1 - self.cycle_mul)))
| pytorch-image-models/timm/scheduler/tanh_lr.py/0 | {
"file_path": "pytorch-image-models/timm/scheduler/tanh_lr.py",
"repo_id": "pytorch-image-models",
"token_count": 1960
} | 191 |
""" Summary utilities
Hacked together by / Copyright 2020 Ross Wightman
"""
import csv
import os
from collections import OrderedDict
try:
import wandb
except ImportError:
pass
def get_outdir(path, *paths, inc=False):
outdir = os.path.join(path, *paths)
if not os.path.exists(outdir):
os.makedirs(outdir)
elif inc:
count = 1
outdir_inc = outdir + '-' + str(count)
while os.path.exists(outdir_inc):
count = count + 1
outdir_inc = outdir + '-' + str(count)
assert count < 100
outdir = outdir_inc
os.makedirs(outdir)
return outdir
def update_summary(
epoch,
train_metrics,
eval_metrics,
filename,
lr=None,
write_header=False,
log_wandb=False,
):
rowd = OrderedDict(epoch=epoch)
rowd.update([('train_' + k, v) for k, v in train_metrics.items()])
if eval_metrics:
rowd.update([('eval_' + k, v) for k, v in eval_metrics.items()])
if lr is not None:
rowd['lr'] = lr
if log_wandb:
wandb.log(rowd)
with open(filename, mode='a') as cf:
dw = csv.DictWriter(cf, fieldnames=rowd.keys())
if write_header: # first iteration (epoch == 1 can't be used)
dw.writeheader()
dw.writerow(rowd)
| pytorch-image-models/timm/utils/summary.py/0 | {
"file_path": "pytorch-image-models/timm/utils/summary.py",
"repo_id": "pytorch-image-models",
"token_count": 633
} | 192 |
# Rust builder
FROM lukemathwalker/cargo-chef:latest-rust-1.71 AS chef
WORKDIR /usr/src
ARG CARGO_REGISTRIES_CRATES_IO_PROTOCOL=sparse
FROM chef as planner
COPY Cargo.toml Cargo.toml
COPY rust-toolchain.toml rust-toolchain.toml
COPY proto proto
COPY benchmark benchmark
COPY router router
COPY launcher launcher
RUN cargo chef prepare --recipe-path recipe.json
FROM chef AS builder
ARG GIT_SHA
ARG DOCKER_LABEL
RUN PROTOC_ZIP=protoc-21.12-linux-x86_64.zip && \
curl -OL https://github.com/protocolbuffers/protobuf/releases/download/v21.12/$PROTOC_ZIP && \
unzip -o $PROTOC_ZIP -d /usr/local bin/protoc && \
unzip -o $PROTOC_ZIP -d /usr/local 'include/*' && \
rm -f $PROTOC_ZIP
COPY --from=planner /usr/src/recipe.json recipe.json
RUN cargo chef cook --release --recipe-path recipe.json
COPY Cargo.toml Cargo.toml
COPY rust-toolchain.toml rust-toolchain.toml
COPY proto proto
COPY benchmark benchmark
COPY router router
COPY launcher launcher
RUN cargo build --release
# Text Generation Inference base image for RoCm
FROM rocm/dev-ubuntu-20.04:5.7 as base
RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
build-essential \
ca-certificates \
ccache \
curl \
git \
make \
libssl-dev \
g++ \
# Needed to build VLLM & flash.
rocthrust-dev \
hipsparse-dev \
hipblas-dev && \
rm -rf /var/lib/apt/lists/*
# Keep in sync with `server/pyproject.toml
ARG MAMBA_VERSION=23.1.0-1
ARG PYTORCH_VERSION='2.2.0.dev0'
ARG ROCM_VERSION='5.7'
ARG PYTHON_VERSION='3.10.10'
# Automatically set by buildx
ARG TARGETPLATFORM
ENV PATH /opt/conda/bin:$PATH
# TGI seem to require libssl.so.1.1 instead of libssl.so.3 so we can't use ubuntu 22.04. Ubuntu 20.04 has python==3.8, and TGI requires python>=3.9, hence the need for miniconda.
# Install mamba
# translating Docker's TARGETPLATFORM into mamba arches
RUN case ${TARGETPLATFORM} in \
"linux/arm64") MAMBA_ARCH=aarch64 ;; \
*) MAMBA_ARCH=x86_64 ;; \
esac && \
curl -fsSL -v -o ~/mambaforge.sh -O "https://github.com/conda-forge/miniforge/releases/download/${MAMBA_VERSION}/Mambaforge-${MAMBA_VERSION}-Linux-${MAMBA_ARCH}.sh"
RUN chmod +x ~/mambaforge.sh && \
bash ~/mambaforge.sh -b -p /opt/conda && \
mamba init && \
rm ~/mambaforge.sh
# Install PyTorch 2.2 RC compiled against RoCm 5.7, as VLLM can not be compiled with RoCm 5.6.
RUN pip install torch --index-url https://download.pytorch.org/whl/test/rocm5.7/
FROM base AS kernel-builder
# Build vllm kernels
FROM kernel-builder AS vllm-builder
WORKDIR /usr/src
COPY server/Makefile-vllm Makefile
# Build specific version of vllm
RUN make build-vllm-rocm
# Build Flash Attention v2 kernels
FROM kernel-builder AS flash-att-v2-builder
WORKDIR /usr/src
COPY server/Makefile-flash-att-v2 Makefile
# Build specific version of flash attention v2
RUN make build-flash-attention-v2-rocm
# Build Transformers CUDA kernels (gpt-neox and bloom)
FROM kernel-builder as custom-kernels-builder
WORKDIR /usr/src
COPY server/custom_kernels/ .
RUN PYTORCH_ROCM_ARCH=gfx90a python setup.py build
# Build exllama kernels
FROM kernel-builder as exllama-kernels-builder
WORKDIR /usr/src
COPY server/exllama_kernels/ .
RUN PYTORCH_ROCM_ARCH="gfx90a" python setup.py build
# Build exllama v2 kernels
FROM kernel-builder as exllamav2-kernels-builder
WORKDIR /usr/src
COPY server/exllamav2_kernels/ .
RUN PYTORCH_ROCM_ARCH="gfx90a" python setup.py build
FROM base as base-copy
# Text Generation Inference base env
ENV HUGGINGFACE_HUB_CACHE=/data \
HF_HUB_ENABLE_HF_TRANSFER=1 \
PORT=80
# Copy builds artifacts from vllm builder
COPY --from=vllm-builder /usr/src/vllm/build/lib.linux-x86_64-cpython-310 /opt/conda/lib/python3.10/site-packages
# Copy build artifacts from flash attention v2 builder
COPY --from=flash-att-v2-builder /usr/src/flash-attention-v2/build/lib.linux-x86_64-cpython-310 /opt/conda/lib/python3.10/site-packages
# Copy build artifacts from custom kernels builder
COPY --from=custom-kernels-builder /usr/src/build/lib.linux-x86_64-cpython-310 /opt/conda/lib/python3.10/site-packages
# Copy build artifacts from exllama kernels builder
COPY --from=exllama-kernels-builder /usr/src/build/lib.linux-x86_64-cpython-310 /opt/conda/lib/python3.10/site-packages
# Copy build artifacts from exllamav2 kernels builder
COPY --from=exllamav2-kernels-builder /usr/src/build/lib.linux-x86_64-cpython-310 /opt/conda/lib/python3.10/site-packages
# Install flash-attention dependencies
RUN pip install einops --no-cache-dir
# Install server
COPY proto proto
COPY server server
COPY server/Makefile server/Makefile
RUN cd server && \
make gen-server && \
pip install -r requirements_rocm.txt && \
pip install ".[accelerate, peft]" --no-cache-dir
# Install benchmarker
COPY --from=builder /usr/src/target/release/text-generation-benchmark /usr/local/bin/text-generation-benchmark
# Install router
COPY --from=builder /usr/src/target/release/text-generation-router /usr/local/bin/text-generation-router
# Install launcher
COPY --from=builder /usr/src/target/release/text-generation-launcher /usr/local/bin/text-generation-launcher
# AWS Sagemaker compatible image
FROM base-copy as sagemaker
COPY sagemaker-entrypoint.sh entrypoint.sh
RUN chmod +x entrypoint.sh
ENTRYPOINT ["./entrypoint.sh"]
# Final image
FROM base-copy
ENTRYPOINT ["text-generation-launcher"]
CMD ["--json-output"]
| text-generation-inference/Dockerfile_amd/0 | {
"file_path": "text-generation-inference/Dockerfile_amd",
"repo_id": "text-generation-inference",
"token_count": 2127
} | 193 |
unit-tests:
python -m pytest --cov=text_generation tests
install:
pip install pip --upgrade
pip install -e . | text-generation-inference/clients/python/Makefile/0 | {
"file_path": "text-generation-inference/clients/python/Makefile",
"repo_id": "text-generation-inference",
"token_count": 40
} | 194 |
- sections:
- local: index
title: Text Generation Inference
- local: quicktour
title: Quick Tour
- local: installation
title: Installation
- local: supported_models
title: Supported Models and Hardware
- local: messages_api
title: Messages API
title: Getting started
- sections:
- local: basic_tutorials/consuming_tgi
title: Consuming TGI
- local: basic_tutorials/preparing_model
title: Preparing Model for Serving
- local: basic_tutorials/gated_model_access
title: Serving Private & Gated Models
- local: basic_tutorials/using_cli
title: Using TGI CLI
- local: basic_tutorials/launcher
title: All TGI CLI options
- local: basic_tutorials/non_core_models
title: Non-core Model Serving
title: Tutorials
- sections:
- local: conceptual/streaming
title: Streaming
- local: conceptual/quantization
title: Quantization
- local: conceptual/tensor_parallelism
title: Tensor Parallelism
- local: conceptual/paged_attention
title: PagedAttention
- local: conceptual/safetensors
title: Safetensors
- local: conceptual/flash_attention
title: Flash Attention
title: Conceptual Guides
| text-generation-inference/docs/source/_toctree.yml/0 | {
"file_path": "text-generation-inference/docs/source/_toctree.yml",
"repo_id": "text-generation-inference",
"token_count": 384
} | 195 |
# Quick Tour
The easiest way of getting started is using the official Docker container. Install Docker following [their installation instructions](https://docs.docker.com/get-docker/).
Let's say you want to deploy [Falcon-7B Instruct](https://huggingface.co/tiiuae/falcon-7b-instruct) model with TGI. Here is an example on how to do that:
```bash
model=tiiuae/falcon-7b-instruct
volume=$PWD/data # share a volume with the Docker container to avoid downloading weights every run
docker run --gpus all --shm-size 1g -p 8080:80 -v $volume:/data ghcr.io/huggingface/text-generation-inference:1.4 --model-id $model
```
<Tip warning={true}>
To use NVIDIA GPUs, you need to install the [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html). We also recommend using NVIDIA drivers with CUDA version 12.2 or higher.
</Tip>
TGI also supports ROCm-enabled AMD GPUs (only MI210 and MI250 are tested), details are available in the [Supported Hardware section](./supported_models#supported-hardware) and [AMD documentation](https://rocm.docs.amd.com/en/latest/deploy/docker.html). To launch TGI on ROCm GPUs, please use instead:
```bash
docker run --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --device=/dev/kfd --device=/dev/dri --group-add video --ipc=host --shm-size 1g -p 8080:80 -v $volume:/data ghcr.io/huggingface/text-generation-inference:1.4-rocm --model-id $model
```
Once TGI is running, you can use the `generate` endpoint by doing requests. To learn more about how to query the endpoints, check the [Consuming TGI](./basic_tutorials/consuming_tgi) section, where we show examples with utility libraries and UIs. Below you can see a simple snippet to query the endpoint.
<inferencesnippet>
<python>
```python
import requests
headers = {
"Content-Type": "application/json",
}
data = {
'inputs': 'What is Deep Learning?',
'parameters': {
'max_new_tokens': 20,
},
}
response = requests.post('http://127.0.0.1:8080/generate', headers=headers, json=data)
print(response.json())
# {'generated_text': '\n\nDeep Learning is a subset of Machine Learning that is concerned with the development of algorithms that can'}
```
</python>
<js>
```js
async function query() {
const response = await fetch(
'http://127.0.0.1:8080/generate',
{
method: 'POST',
headers: { 'Content-Type': 'application/json'},
body: JSON.stringify({
'inputs': 'What is Deep Learning?',
'parameters': {
'max_new_tokens': 20
}
})
}
);
}
query().then((response) => {
console.log(JSON.stringify(response));
});
/// {"generated_text":"\n\nDeep Learning is a subset of Machine Learning that is concerned with the development of algorithms that can"}
```
</js>
<curl>
```curl
curl 127.0.0.1:8080/generate \
-X POST \
-d '{"inputs":"What is Deep Learning?","parameters":{"max_new_tokens":20}}' \
-H 'Content-Type: application/json'
```
</curl>
</inferencesnippet>
<Tip>
To see all possible deploy flags and options, you can use the `--help` flag. It's possible to configure the number of shards, quantization, generation parameters, and more.
```bash
docker run ghcr.io/huggingface/text-generation-inference:1.4 --help
```
</Tip>
| text-generation-inference/docs/source/quicktour.md/0 | {
"file_path": "text-generation-inference/docs/source/quicktour.md",
"repo_id": "text-generation-inference",
"token_count": 1205
} | 196 |
{
"details": {
"best_of_sequences": null,
"finish_reason": "length",
"generated_tokens": 10,
"prefill": [
{
"id": 1,
"logprob": null,
"text": "<s>"
},
{
"id": 4321,
"logprob": -8.6875,
"text": "Test"
},
{
"id": 2009,
"logprob": -11.546875,
"text": "request"
}
],
"seed": null,
"tokens": [
{
"id": 363,
"logprob": -1.5351562,
"special": false,
"text": " for"
},
{
"id": 847,
"logprob": -2.5722656,
"special": false,
"text": " /"
},
{
"id": 2754,
"logprob": -2.2714844,
"special": false,
"text": "api"
},
{
"id": 29914,
"logprob": -0.03414917,
"special": false,
"text": "/"
},
{
"id": 29894,
"logprob": -0.95996094,
"special": false,
"text": "v"
},
{
"id": 29896,
"logprob": -0.3635254,
"special": false,
"text": "1"
},
{
"id": 29914,
"logprob": -0.013031006,
"special": false,
"text": "/"
},
{
"id": 16418,
"logprob": -3.1523438,
"special": false,
"text": "projects"
},
{
"id": 29914,
"logprob": -0.43701172,
"special": false,
"text": "/"
},
{
"id": 29896,
"logprob": -1.9394531,
"special": false,
"text": "1"
}
],
"top_tokens": null
},
"generated_text": " for /api/v1/projects/1"
}
| text-generation-inference/integration-tests/models/__snapshots__/test_flash_llama/test_flash_llama.json/0 | {
"file_path": "text-generation-inference/integration-tests/models/__snapshots__/test_flash_llama/test_flash_llama.json",
"repo_id": "text-generation-inference",
"token_count": 1050
} | 197 |
{
"details": {
"best_of_sequences": null,
"finish_reason": "length",
"generated_tokens": 10,
"prefill": [
{
"id": 14402,
"logprob": null,
"text": "Test"
},
{
"id": 2581,
"logprob": -11.6171875,
"text": " request"
}
],
"seed": null,
"tokens": [
{
"id": 25,
"logprob": -2.3203125,
"special": false,
"text": ":"
},
{
"id": 1391,
"logprob": -0.98779297,
"special": false,
"text": " {"
},
{
"id": 25927,
"logprob": -0.76660156,
"special": false,
"text": "request"
},
{
"id": 92,
"logprob": -0.7246094,
"special": false,
"text": "}"
},
{
"id": 4943,
"logprob": -0.41333008,
"special": false,
"text": "\")"
},
{
"id": 198,
"logprob": -0.11785889,
"special": false,
"text": "\n"
},
{
"id": 50280,
"logprob": -0.97265625,
"special": false,
"text": " "
},
{
"id": 26209,
"logprob": -1.4414062,
"special": false,
"text": "response"
},
{
"id": 796,
"logprob": -0.0569458,
"special": false,
"text": " ="
},
{
"id": 2116,
"logprob": -1.1533203,
"special": false,
"text": " self"
}
],
"top_tokens": null
},
"generated_text": ": {request}\")\n response = self"
}
| text-generation-inference/integration-tests/models/__snapshots__/test_flash_phi/test_flash_phi.json/0 | {
"file_path": "text-generation-inference/integration-tests/models/__snapshots__/test_flash_phi/test_flash_phi.json",
"repo_id": "text-generation-inference",
"token_count": 1003
} | 198 |
{
"details": {
"best_of_sequences": null,
"finish_reason": "eos_token",
"generated_tokens": 9,
"prefill": [
{
"id": 0,
"logprob": null,
"text": "<pad>"
}
],
"seed": 0,
"tokens": [
{
"id": 16017,
"logprob": -0.30908203,
"special": false,
"text": " blue"
},
{
"id": 20495,
"logprob": 0.0,
"special": false,
"text": " sky"
},
{
"id": 259,
"logprob": -0.28271484,
"special": false,
"text": " "
},
{
"id": 15484,
"logprob": -1.7929688,
"special": false,
"text": "appear"
},
{
"id": 345,
"logprob": -0.8935547,
"special": false,
"text": "ed"
},
{
"id": 281,
"logprob": 0.0,
"special": false,
"text": " in"
},
{
"id": 287,
"logprob": 0.0,
"special": false,
"text": " the"
},
{
"id": 20495,
"logprob": -0.32299805,
"special": false,
"text": " sky"
},
{
"id": 1,
"logprob": 0.0,
"special": true,
"text": "</s>"
}
]
},
"generated_text": "Why is the sky blue?blue sky appeared in the sky"
}
| text-generation-inference/integration-tests/models/__snapshots__/test_mt0_base/test_mt0_base_all_params.json/0 | {
"file_path": "text-generation-inference/integration-tests/models/__snapshots__/test_mt0_base/test_mt0_base_all_params.json",
"repo_id": "text-generation-inference",
"token_count": 831
} | 199 |
import pytest
@pytest.fixture(scope="module")
def flash_mistral_handle(launcher):
with launcher("mistralai/Mistral-7B-Instruct-v0.1") as handle:
yield handle
@pytest.fixture(scope="module")
async def flash_mistral(flash_mistral_handle):
await flash_mistral_handle.health(300)
return flash_mistral_handle.client
@pytest.mark.asyncio
@pytest.mark.private
async def test_flash_mistral(flash_mistral, response_snapshot):
response = await flash_mistral.generate(
"Test request", max_new_tokens=10, decoder_input_details=True
)
assert response.details.generated_tokens == 10
assert response.generated_text == ": Let n = 10 - 1"
assert response == response_snapshot
@pytest.mark.asyncio
@pytest.mark.private
async def test_flash_mistral_all_params(flash_mistral, response_snapshot):
response = await flash_mistral.generate(
"Test request",
max_new_tokens=10,
repetition_penalty=1.2,
return_full_text=True,
stop_sequences=["test"],
temperature=0.5,
top_p=0.9,
top_k=10,
truncate=5,
typical_p=0.9,
watermark=True,
decoder_input_details=True,
seed=0,
)
assert response.details.generated_tokens == 10
assert response == response_snapshot
@pytest.mark.asyncio
@pytest.mark.private
async def test_flash_mistral_load(flash_mistral, generate_load, response_snapshot):
responses = await generate_load(
flash_mistral, "Test request", max_new_tokens=10, n=4
)
assert len(responses) == 4
assert all(
[r.generated_text == responses[0].generated_text for r in responses]
), f"{[r.generated_text for r in responses]}"
assert responses[0].generated_text == ": Let n = 10 - 1"
assert responses == response_snapshot
| text-generation-inference/integration-tests/models/test_flash_mistral.py/0 | {
"file_path": "text-generation-inference/integration-tests/models/test_flash_mistral.py",
"repo_id": "text-generation-inference",
"token_count": 738
} | 200 |
//! Text Generation gRPC client library
mod client;
#[allow(clippy::derive_partial_eq_without_eq)]
mod pb;
mod sharded_client;
pub use client::Client;
pub use pb::generate::v2::HealthResponse;
pub use pb::generate::v2::InfoResponse as ShardInfo;
pub use pb::generate::v2::{
Batch, CachedBatch, FinishReason, GeneratedText, Generation, NextTokenChooserParameters,
Request, StoppingCriteriaParameters, Tokens,
};
pub use sharded_client::ShardedClient;
use thiserror::Error;
use tonic::transport;
use tonic::Status;
#[derive(Error, Debug, Clone)]
pub enum ClientError {
#[error("Could not connect to Text Generation server: {0}")]
Connection(String),
#[error("Server error: {0}")]
Generation(String),
#[error("Sharded results are empty")]
EmptyResults,
}
impl From<Status> for ClientError {
fn from(err: Status) -> Self {
let err = Self::Generation(err.message().to_string());
tracing::error!("{err}");
err
}
}
impl From<transport::Error> for ClientError {
fn from(err: transport::Error) -> Self {
let err = Self::Connection(err.to_string());
tracing::error!("{err}");
err
}
}
pub type Result<T> = std::result::Result<T, ClientError>;
| text-generation-inference/router/client/src/lib.rs/0 | {
"file_path": "text-generation-inference/router/client/src/lib.rs",
"repo_id": "text-generation-inference",
"token_count": 460
} | 201 |
awq_commit := f084f40bd996f3cf3a0633c1ad7d9d476c318aaa
awq:
rm -rf llm-awq
git clone https://github.com/mit-han-lab/llm-awq
build-awq: awq
cd llm-awq/ && git fetch && git checkout $(awq_commit)
cd llm-awq/awq/kernels && python setup.py build
install-awq: build-awq
pip uninstall awq_inference_engine -y || true
cd llm-awq/awq/kernels && python setup.py install
| text-generation-inference/server/Makefile-awq/0 | {
"file_path": "text-generation-inference/server/Makefile-awq",
"repo_id": "text-generation-inference",
"token_count": 165
} | 202 |
// Adapted from turboderp exllama: https://github.com/turboderp/exllama
#include "q4_matrix.cuh"
#include <vector>
#include "../util.cuh"
#include "../matrix.cuh"
using namespace std;
const int UNSHUF_BLOCKSIZE_X = 64;
const int RECONS_THREADS_X = 64; // Block size and thread count along columns in out, each thread converts 1 column
const int RECONS_THREADS_Y = 1; // Block size and thread count along rows in x and out, each thread converts 8 rows
vector<Q4Matrix*> g_q4_matrices;
void g_q4_keep_matrix(Q4Matrix* m)
{
g_q4_matrices.push_back(m);
}
void g_q4_free_matrices()
{
for (const auto& m : g_q4_matrices) delete m;
g_q4_matrices.clear();
}
Q4Matrix::Q4Matrix
(
const int _height,
const int _width,
const int _groups,
uint32_t* _qweight,
uint32_t* _qzeros,
half* _scales,
uint32_t* _g_idx,
const int _device
) :
height(_height),
width(_width),
groups(_groups),
device(_device)
{
cudaSetDevice(device);
cuda_qweight = _qweight;
cuda_qzeros = _qzeros;
cuda_scales = _scales;
groupsize = height / groups;
if (_g_idx) make_sequential(_g_idx);
}
Q4Matrix::~Q4Matrix()
{
}
// Make sequential
__global__ void make_sequential_kernel
(
const uint32_t* __restrict__ w,
uint32_t* __restrict__ w_new,
const uint32_t* __restrict__ x_map,
const int w_height,
const int w_width
)
{
const uint64_t* w2 = (uint64_t*) w;
uint64_t* w_new2 = (uint64_t*) w_new;
int w2_stride = w_width >> 1;
int w2_column = UNSHUF_BLOCKSIZE_X * blockIdx.x + threadIdx.x;
int w_new2_row = blockIdx.y;
int x_map_idx = w_new2_row << 3;
uint64_t dst = 0;
#pragma unroll
for (int i = 0; i < 8; i++)
{
int source_row = x_map[x_map_idx++];
int w2_row = source_row >> 3;
int w2_subrow = source_row & 0x07;
int w2_row_shift = w2_subrow << 2;
int wnew2_row_shift = i << 2;
uint64_t src = w2[w2_row * w2_stride + w2_column];
src >>= w2_row_shift;
src &= 0x0000000f0000000f;
src <<= wnew2_row_shift;
dst |= src;
}
w_new2[w_new2_row * w2_stride + w2_column] = dst;
}
void Q4Matrix::make_sequential(const uint32_t* cpu_g_idx)
{
uint32_t* cuda_new_qweight = NULL;
cudaMalloc(&cuda_new_qweight, height / 8 * width * sizeof(uint32_t));
cudaMalloc(&cuda_x_map, height * sizeof(uint32_t)); // TODO: Should probably be allocated in PyTorch
uint32_t* cpu_g_idx_map = (uint32_t*) calloc(groups, sizeof(uint32_t));
uint32_t* cpu_x_map = (uint32_t*) malloc(height * sizeof(uint32_t));
uint32_t* cpu_x_map_inv = (uint32_t*) malloc(height * sizeof(uint32_t));
// Group histogram
for (int i = 0; i < height; i++) cpu_g_idx_map[cpu_g_idx[i]]++;
// Group map
for (int i = 0, acc = 0; i < groups; i++)
{
short tmp = cpu_g_idx_map[i];
cpu_g_idx_map[i] = acc;
acc += tmp;
}
// X map (inverse)
for (int row = 0; row < height; row++)
{
uint32_t target_group = cpu_g_idx[row];
uint32_t target_row = cpu_g_idx_map[target_group];
cpu_g_idx_map[target_group]++;
cpu_x_map_inv[row] = target_row;
}
// X map
for (int row = 0; row < height; row++) cpu_x_map[cpu_x_map_inv[row]] = row;
// Move to CUDA
cudaMemcpyAsync(cuda_x_map, cpu_x_map, height * sizeof(uint32_t), cudaMemcpyHostToDevice);
// Rearrange rows in w
dim3 threads(UNSHUF_BLOCKSIZE_X, 1, 1);
dim3 blocks(width / UNSHUF_BLOCKSIZE_X / 2, height / 8, 1);
make_sequential_kernel<<<blocks, threads>>>(cuda_qweight, cuda_new_qweight, cuda_x_map, height / 8, width);
// Replace qweights
cudaMemcpyAsync(cuda_qweight, cuda_new_qweight, height / 8 * width * sizeof(uint32_t), cudaMemcpyDeviceToDevice);
// Cleanup
cudaDeviceSynchronize();
cudaFree(cuda_new_qweight);
free(cpu_g_idx_map);
free(cpu_x_map);
free(cpu_x_map_inv);
}
__global__ void reconstruct_kernel
(
const uint32_t* __restrict__ w,
half* __restrict__ out, // (y)
const half* __restrict__ w_scales,
const uint32_t* __restrict__ w_zeros,
const int height,
const int width,
const int groupsize
)
{
// Start of block
int column = RECONS_THREADS_X * blockIdx.x + threadIdx.x;
int row = (RECONS_THREADS_Y * blockIdx.y + threadIdx.y) * 8;
// Views
MatrixView_q4_column w_(w, height, width);
MatrixView_half_rw out_(out, height, width);
MatrixView_half w_scales_(w_scales, height / groupsize, width);
MatrixView_q4_row w_zeros_(w_zeros, height / groupsize, width);
// Groupsize version
int group = row / groupsize;
half w_scale = w_scales_.item(group, column);
uint32_t w_zero = w_zeros_.item(group, column) + 1;
uint32_t w_read = w_.item_uint32_t(row, column);
half* out_ptr = out_.item_ptr(row, column);
#pragma unroll
for (int s = 0; s < 32; s += 4)
{
half w_item = __hmul(__int2half_rn((int)((w_read >> s) & 0x0f) - w_zero), w_scale);
*out_ptr = w_item; out_ptr += out_.width;
}
}
void Q4Matrix::reconstruct(half* out)
{
dim3 threads(RECONS_THREADS_X, RECONS_THREADS_Y, 1);
dim3 blocks
(
(width + threads.x - 1) / threads.x,
(height / 8 + threads.y - 1) / threads.y,
1
);
reconstruct_kernel<<<blocks, threads>>>(cuda_qweight, out, cuda_scales, cuda_qzeros, height / 8, width, groupsize);
} | text-generation-inference/server/exllama_kernels/exllama_kernels/cuda_func/q4_matrix.cu/0 | {
"file_path": "text-generation-inference/server/exllama_kernels/exllama_kernels/cuda_func/q4_matrix.cu",
"repo_id": "text-generation-inference",
"token_count": 2516
} | 203 |
#include "q_matrix.cuh"
#include "matrix_view.cuh"
#include "util.cuh"
#include "quant/qdq_2.cuh"
#include "quant/qdq_3.cuh"
#include "quant/qdq_4.cuh"
#include "quant/qdq_5.cuh"
#include "quant/qdq_6.cuh"
#include "quant/qdq_8.cuh"
#define BLOCK_KN_SIZE 128
#define THREADS_X 32
#define THREADS_Y 32
// Shuffle quantized data on load
__global__ void shuffle_kernel
(
uint32_t* __restrict__ b_q_weight,
const int size_k,
const int size_n,
const int rows_8,
const int rows_6,
const int rows_5,
const int rows_4,
const int rows_3,
const int rows_2
)
{
int n = blockIdx.x * THREADS_X + threadIdx.x;
if (n >= size_n) return;
int k = 0;
uint32_t* b_ptr = b_q_weight + n;
while (k < rows_8) { shuffle_8bit_4 (b_ptr, size_n); b_ptr += 1 * size_n; k += 4; }
while (k < rows_6) { shuffle_6bit_16(b_ptr, size_n); b_ptr += 3 * size_n; k += 16; }
while (k < rows_5) { shuffle_5bit_32(b_ptr, size_n); b_ptr += 5 * size_n; k += 32; }
while (k < rows_4) { shuffle_4bit_8 (b_ptr, size_n); b_ptr += 1 * size_n; k += 8; }
while (k < rows_3) { shuffle_3bit_32(b_ptr, size_n); b_ptr += 3 * size_n; k += 32; }
while (k < rows_2) { shuffle_2bit_16(b_ptr, size_n); b_ptr += 1 * size_n; k += 16; }
}
// QMatrix constructor
QMatrix::QMatrix
(
const int _device,
const int _height,
const int _width,
const int _groups,
uint32_t* _q_weight,
uint16_t* _q_perm,
uint16_t* _q_invperm,
uint32_t* _q_scale,
half* _q_scale_max,
uint16_t* _q_groups,
uint16_t* _q_group_map,
uint32_t* _gptq_qzeros,
half* _gptq_scales,
uint32_t* _gptq_g_idx,
half* _temp_dq
) :
device(_device),
height(_height),
width(_width),
groups(_groups),
temp_dq(_temp_dq)
{
cudaSetDevice(device);
failed = false;
cuda_q_weight = _q_weight;
cuda_q_perm = _q_perm;
cuda_q_invperm = _q_invperm;
cuda_q_scale = _q_scale;
cuda_q_scale_max = _q_scale_max;
cuda_q_groups = _q_groups;
cuda_q_group_map = _q_group_map;
cuda_gptq_qzeros = _gptq_qzeros;
cuda_gptq_scales = _gptq_scales;
is_gptq = (_gptq_qzeros != NULL);
if (is_gptq)
{
gptq_groupsize = 1;
while (gptq_groupsize * groups < height) gptq_groupsize *= 2;
}
// Create group map
rows_8 = 0;
rows_6 = 0;
rows_5 = 0;
rows_4 = 0;
rows_3 = 0;
rows_2 = 0;
if (!is_gptq)
{
uint16_t* cpu_q_groups = (uint16_t*)calloc(groups * 2, sizeof(uint16_t));
cudaMemcpy(cpu_q_groups, cuda_q_groups, groups * 2 * sizeof(uint16_t), cudaMemcpyDeviceToHost);
int row = 0;
for (int i = 0; i < groups; i++)
{
int bits = cpu_q_groups[i * 2];
int rows;
if (i < groups - 1)
{
int qrows = cpu_q_groups[i * 2 + 3] - cpu_q_groups[i * 2 + 1];
rows = qrows * 32 / bits;
}
else rows = height - row;
if (bits == 8) rows_8 += rows;
if (bits == 6) rows_6 += rows;
if (bits == 5) rows_5 += rows;
if (bits == 4) rows_4 += rows;
if (bits == 3) rows_3 += rows;
if (bits == 2) rows_2 += rows;
row += rows;
}
free(cpu_q_groups);
rows_6 += rows_8;
rows_5 += rows_6;
rows_4 += rows_5;
rows_3 += rows_4;
rows_2 += rows_3;
}
else
{
rows_4 = height;
rows_3 = height;
rows_2 = height;
if (_gptq_g_idx)
{
if (!make_sequential(_gptq_g_idx))
{
failed = true;
//printf("FAIL\n");
return;
}
}
}
// DBGI(rows_8);
// DBGI(rows_6);
// DBGI(rows_5);
// DBGI(rows_4);
// DBGI(rows_3);
// DBGI(rows_2);
// Shuffle quantized data
dim3 blockDim, gridDim;
blockDim.x = THREADS_X;
blockDim.y = 1;
gridDim.x = DIVIDE(width, THREADS_X);
gridDim.y = 1;
shuffle_kernel<<<gridDim, blockDim>>>(cuda_q_weight, height, width, rows_8, rows_6, rows_5, rows_4, rows_3, rows_2);
}
QMatrix::~QMatrix()
{
}
// Reconstruct b[k,n] (GPTQ)
__global__ void reconstruct_gptq_kernel
(
const uint32_t* __restrict__ b_q_weight,
const uint16_t* __restrict__ b_q_perm,
const uint32_t* __restrict__ b_gptq_qzeros,
const half* __restrict__ b_gptq_scales,
//const uint16_t* __restrict__ b_q_groups,
const int size_k,
const int size_n,
const int groupsize,
const int groups,
half* __restrict__ b,
const int rows_4
)
{
MatrixView_half_rw b_(b, size_k, size_n);
MatrixView_q4_row b_gptq_qzeros_(b_gptq_qzeros, groups, size_n);
MatrixView_half b_gptq_scales_(b_gptq_scales, groups, size_n);
int offset_k = BLOCK_KN_SIZE * blockIdx.y;
int offset_n = BLOCK_KN_SIZE * blockIdx.x * 4;
int end_k = min(offset_k + BLOCK_KN_SIZE, size_k);
// Preload remapping table
__shared__ uint16_t perm[BLOCK_KN_SIZE];
int t = threadIdx.x;
if (b_q_perm)
{
if (offset_k + t < size_k)
perm[t] = b_q_perm[offset_k + t];
}
// Column
int n = offset_n + t * 4;
if (n >= size_n) return;
// Find initial group
int group = offset_k / groupsize;
int nextgroup = offset_k + groupsize;
// b offset
int qk = offset_k / (32 / 4);
const uint32_t* b_ptr = b_q_weight + qk * size_n + n;
// Initial zeros/scale
int zeros[4];
half2 scales[4];
half2 z1z16[4][2];
half2 y1y16[4][2];
b_gptq_qzeros_.item4(zeros, group, n);
b_gptq_scales_.item4_h2(scales, group, n);
dequant_4bit_8_prep_zero(zeros[0] + 1, z1z16[0], y1y16[0]);
dequant_4bit_8_prep_zero(zeros[1] + 1, z1z16[1], y1y16[1]);
dequant_4bit_8_prep_zero(zeros[2] + 1, z1z16[2], y1y16[2]);
dequant_4bit_8_prep_zero(zeros[3] + 1, z1z16[3], y1y16[3]);
__syncthreads();
int k = offset_k;
int lk = 0;
while (k < end_k)
{
if (k == nextgroup)
{
group++;
nextgroup += groupsize;
b_gptq_qzeros_.item4(zeros, group, n);
b_gptq_scales_.item4_h2(scales, group, n);
dequant_4bit_8_prep_zero(zeros[0] + 1, z1z16[0], y1y16[0]);
dequant_4bit_8_prep_zero(zeros[1] + 1, z1z16[1], y1y16[1]);
dequant_4bit_8_prep_zero(zeros[2] + 1, z1z16[2], y1y16[2]);
dequant_4bit_8_prep_zero(zeros[3] + 1, z1z16[3], y1y16[3]);
}
for (int p = 0; p < 4; p++)
{
half2 dq[4][4];
const int4* b_ptr4 = (int4*) b_ptr;
int4 load_int4 = *b_ptr4;
dequant_4bit_8_gptq(load_int4.x, dq[0], z1z16[0], y1y16[0], size_n, false);
dequant_4bit_8_gptq(load_int4.y, dq[1], z1z16[1], y1y16[1], size_n, false);
dequant_4bit_8_gptq(load_int4.z, dq[2], z1z16[2], y1y16[2], size_n, false);
dequant_4bit_8_gptq(load_int4.w, dq[3], z1z16[3], y1y16[3], size_n, false);
b_ptr += size_n;
//half* dqh = (half*)dq;
if (b_q_perm)
{
for (int j = 0; j < 4; j++)
{
for (int v = 0; v < 4; v++) dq[v][j] = __hmul2(scales[v], dq[v][j]);
b_.set4(perm[lk++], n, __low2half(dq[0][j]), __low2half(dq[1][j]), __low2half(dq[2][j]), __low2half(dq[3][j]));
b_.set4(perm[lk++], n, __high2half(dq[0][j]), __high2half(dq[1][j]), __high2half(dq[2][j]), __high2half(dq[3][j]));
}
}
else
{
for (int j = 0; j < 4; j++)
{
for (int v = 0; v < 4; v++) dq[v][j] = __hmul2(scales[v], dq[v][j]);
b_.set4(offset_k + lk++, n, __low2half(dq[0][j]), __low2half(dq[1][j]), __low2half(dq[2][j]), __low2half(dq[3][j]));
b_.set4(offset_k + lk++, n, __high2half(dq[0][j]), __high2half(dq[1][j]), __high2half(dq[2][j]), __high2half(dq[3][j]));
}
}
}
k += 32;
}
}
// Reconstruct b[k,n]
__global__ void reconstruct_kernel
(
const uint32_t* __restrict__ b_q_weight,
const uint16_t* __restrict__ b_q_perm,
const uint32_t* __restrict__ b_q_scale,
const half* __restrict__ b_q_scale_max,
const uint16_t* __restrict__ b_q_group_map,
const int size_k,
const int size_n,
//const int groupsize,
const int groups,
half* __restrict__ b,
const int rows_8,
const int rows_6,
const int rows_5,
const int rows_4,
const int rows_3,
const int rows_2
)
{
MatrixView_half_rw b_(b, size_k, size_n);
MatrixView_q4_row b_q_scale_(b_q_scale, groups, size_n);
int offset_k = BLOCK_KN_SIZE * blockIdx.y;
int offset_n = BLOCK_KN_SIZE * blockIdx.x;
// Preload remapping table
int t = threadIdx.x;
__shared__ uint16_t perm[BLOCK_KN_SIZE];
if (offset_k + t < size_k)
perm[t] = b_q_perm[offset_k + t];
// Column
int n = offset_n + t;
if (n >= size_n) return;
// Find initial group
// int group = offset_k / groupsize;
int group = b_q_group_map[offset_k * 2];
int pre_rows_8 = min(rows_8, offset_k);
int pre_rows_6 = offset_k > rows_8 ? min(rows_6, offset_k) - rows_8 : 0;
int pre_rows_5 = offset_k > rows_6 ? min(rows_5, offset_k) - rows_6 : 0;
int pre_rows_4 = offset_k > rows_5 ? min(rows_4, offset_k) - rows_5 : 0;
int pre_rows_3 = offset_k > rows_4 ? min(rows_3, offset_k) - rows_4 : 0;
int pre_rows_2 = offset_k > rows_3 ? min(rows_2, offset_k) - rows_3 : 0;
int qk = 0;
qk += pre_rows_8 / 32 * 8;
qk += pre_rows_6 / 32 * 6;
qk += pre_rows_5 / 32 * 5;
qk += pre_rows_4 / 32 * 4;
qk += pre_rows_3 / 32 * 3;
qk += pre_rows_2 / 32 * 2;
const uint32_t* b_ptr = b_q_weight + qk * size_n + n;
half qs_h = dq_scale(b_q_scale_.item(group, n), b_q_scale_max[group]);
half2 qs_h2 = __halves2half2(qs_h, qs_h);
int nextgroup = offset_k + b_q_group_map[offset_k * 2 + 1];
int end_k = min(offset_k + BLOCK_KN_SIZE, size_k);
int k = offset_k;
int lk = 0;
__syncthreads();
while (k < rows_8 && k < end_k)
{
if (k == nextgroup) { group++; qs_h = dq_scale(b_q_scale_.item(group, n), b_q_scale_max[group]); nextgroup += b_q_group_map[k * 2 + 1]; qs_h2 = __halves2half2(qs_h, qs_h); }
for (int p = 0; p < 4; p++)
{
half2 dq[4];
uint32_t q_0 = *b_ptr; b_ptr += size_n;
uint32_t q_1 = *b_ptr; b_ptr += size_n;
dequant_8bit_8(q_0, q_1, dq, size_n);
for (int j = 0; j < 4; j++) dq[j] = __hmul2(dq[j], qs_h2);
half* dqh = (half*) dq;
for (int j = 0; j < 8; j++) b_.set(perm[lk++], n, dqh[j]);
}
k += 32;
}
while (k < rows_6 && k < end_k)
{
if (k == nextgroup) { group++; qs_h = dq_scale(b_q_scale_.item(group, n), b_q_scale_max[group]); nextgroup += b_q_group_map[k * 2 + 1]; qs_h2 = __halves2half2(qs_h, qs_h); }
for (int p = 0; p < 2; p++)
{
half2 dq[8];
uint32_t q_0 = *b_ptr; b_ptr += size_n;
uint32_t q_1 = *b_ptr; b_ptr += size_n;
uint32_t q_2 = *b_ptr; b_ptr += size_n;
dequant_6bit_16(q_0, q_1, q_2, dq, size_n);
for (int j = 0; j < 8; j++) dq[j] = __hmul2(dq[j], qs_h2);
half* dqh = (half*) dq;
for (int j = 0; j < 16; j++) b_.set(perm[lk++], n, dqh[j]);
}
k += 32;
}
while (k < rows_5 && k < end_k)
{
if (k == nextgroup) { group++; qs_h = dq_scale(b_q_scale_.item(group, n), b_q_scale_max[group]); nextgroup += b_q_group_map[k * 2 + 1]; qs_h2 = __halves2half2(qs_h, qs_h); }
for (int p = 0; p < 1; p++)
{
half2 dq[16];
uint32_t q_0 = *b_ptr; b_ptr += size_n;
uint32_t q_1 = *b_ptr; b_ptr += size_n;
uint32_t q_2 = *b_ptr; b_ptr += size_n;
uint32_t q_3 = *b_ptr; b_ptr += size_n;
uint32_t q_4 = *b_ptr; b_ptr += size_n;
dequant_5bit_32(q_0, q_1, q_2, q_3, q_4, dq, size_n);
for (int j = 0; j < 16; j++) dq[j] = __hmul2(dq[j], qs_h2);
half* dqh = (half*) dq;
for (int j = 0; j < 32; j++) b_.set(perm[lk++], n, dqh[j]);
}
k += 32;
}
while (k < rows_4 && k < end_k)
{
if (k == nextgroup) { group++; qs_h = dq_scale(b_q_scale_.item(group, n), b_q_scale_max[group]); nextgroup += b_q_group_map[k * 2 + 1]; qs_h2 = __halves2half2(qs_h, qs_h); }
for (int p = 0; p < 4; p++)
{
half2 dq[4];
uint32_t q_0 = *b_ptr; b_ptr += size_n;
dequant_4bit_8(q_0, dq, size_n);
for (int j = 0; j < 4; j++) dq[j] = __hmul2(dq[j], qs_h2);
half* dqh = (half*) dq;
for (int j = 0; j < 8; j++) b_.set(perm[lk++], n, dqh[j]);
}
k += 32;
}
while (k < rows_3 && k < end_k)
{
if (k == nextgroup) { group++; qs_h = dq_scale(b_q_scale_.item(group, n), b_q_scale_max[group]); nextgroup += b_q_group_map[k * 2 + 1]; qs_h2 = __halves2half2(qs_h, qs_h); }
for (int p = 0; p < 1; p++)
{
half2 dq[16];
uint32_t q_0 = *b_ptr; b_ptr += size_n;
uint32_t q_1 = *b_ptr; b_ptr += size_n;
uint32_t q_2 = *b_ptr; b_ptr += size_n;
dequant_3bit_32(q_0, q_1, q_2, dq, size_n);
for (int j = 0; j < 16; j++) dq[j] = __hmul2(dq[j], qs_h2);
half* dqh = (half*) dq;
for (int j = 0; j < 32; j++) b_.set(perm[lk++], n, dqh[j]);
}
k += 32;
}
while (k < rows_2 && k < end_k)
{
if (k == nextgroup) { group++; qs_h = dq_scale(b_q_scale_.item(group, n), b_q_scale_max[group]); nextgroup += b_q_group_map[k * 2 + 1]; qs_h2 = __halves2half2(qs_h, qs_h); }
for (int p = 0; p < 1; p++)
{
half2 dq[8];
uint32_t q_0 = *b_ptr; b_ptr += size_n;
dequant_2bit_16(q_0, dq, size_n);
for (int j = 0; j < 8; j++) dq[j] = __hmul2(dq[j], qs_h2);
half* dqh = (half*) dq;
for (int j = 0; j < 16; j++) b_.set(perm[lk++], n, dqh[j]);
}
k += 16;
}
}
void QMatrix::reconstruct(half* out)
{
dim3 blockDim, gridDim;
blockDim.x = BLOCK_KN_SIZE;
blockDim.y = 1;
gridDim.y = DIVIDE(height, BLOCK_KN_SIZE);
if (!is_gptq)
{
gridDim.x = DIVIDE(width, BLOCK_KN_SIZE);
reconstruct_kernel<<<gridDim, blockDim>>>
(
cuda_q_weight,
cuda_q_perm,
cuda_q_scale,
cuda_q_scale_max,
cuda_q_group_map,
height,
width,
//groupsize,
groups,
out,
rows_8,
rows_6,
rows_5,
rows_4,
rows_3,
rows_2
);
}
else
{
gridDim.x = DIVIDE(width, BLOCK_KN_SIZE * 4);
reconstruct_gptq_kernel<<<gridDim, blockDim>>>
(
cuda_q_weight,
cuda_q_perm,
cuda_gptq_qzeros,
cuda_gptq_scales,
//const uint16_t* __restrict__ b_q_groups,
height,
width,
gptq_groupsize,
groups,
out,
rows_4
);
}
}
__global__ void make_sequential_kernel
(
const uint32_t* __restrict__ w,
uint32_t* __restrict__ w_new,
const uint16_t* __restrict__ q_perm,
const int w_height,
const int w_width
)
{
const uint64_t* w2 = (uint64_t*) w;
uint64_t* w_new2 = (uint64_t*) w_new;
int w2_stride = w_width >> 1;
int w2_column = THREADS_X * blockIdx.x + threadIdx.x;
if (w2_column >= w2_stride) return;
int w_new2_row = blockIdx.y;
int q_perm_idx = w_new2_row << 3;
uint64_t dst = 0;
#pragma unroll
for (int i = 0; i < 8; i++)
{
int source_row = q_perm[q_perm_idx++];
int w2_row = source_row >> 3;
int w2_subrow = source_row & 0x07;
int w2_row_shift = w2_subrow << 2;
int wnew2_row_shift = i << 2;
uint64_t src = w2[w2_row * w2_stride + w2_column];
src >>= w2_row_shift;
src &= 0x0000000f0000000f;
src <<= wnew2_row_shift;
dst |= src;
}
w_new2[w_new2_row * w2_stride + w2_column] = dst;
}
bool QMatrix::make_sequential(const uint32_t* cpu_g_idx)
{
uint32_t* cuda_new_qweight = NULL;
cudaError_t err = cudaMalloc(&cuda_new_qweight, height / 8 * width * sizeof(uint32_t));
if (err != cudaSuccess) {
cudaError_t cuda_status = cudaGetLastError(); // Clear error
return false;
}
uint32_t* cpu_g_idx_map = (uint32_t*) calloc(groups, sizeof(uint32_t));
uint32_t* cpu_x_map = (uint32_t*) malloc(height * sizeof(uint32_t));
uint32_t* cpu_x_map_inv = (uint32_t*) malloc(height * sizeof(uint32_t));
// Group histogram
for (int i = 0; i < height; i++) cpu_g_idx_map[cpu_g_idx[i]]++;
// Group map
for (int i = 0, acc = 0; i < groups; i++)
{
short tmp = cpu_g_idx_map[i];
cpu_g_idx_map[i] = acc;
acc += tmp;
}
// X map (inverse)
for (int row = 0; row < height; row++)
{
uint32_t target_group = cpu_g_idx[row];
uint32_t target_row = cpu_g_idx_map[target_group];
cpu_g_idx_map[target_group]++;
cpu_x_map_inv[row] = target_row;
}
// X map
for (int row = 0; row < height; row++) cpu_x_map[cpu_x_map_inv[row]] = row;
// Reduce to uint16_t
uint16_t* cpu_x_map16 = (uint16_t*)cpu_x_map;
uint16_t* cpu_x_map_inv16 = (uint16_t*)cpu_x_map_inv;
for (int row = 0; row < height; row++) cpu_x_map16[row] = (uint16_t) cpu_x_map[row];
for (int row = 0; row < height; row++) cpu_x_map_inv16[row] = (uint16_t) cpu_x_map_inv[row];
// Move to CUDA
cudaMemcpyAsync(cuda_q_perm, cpu_x_map16, height * sizeof(uint16_t), cudaMemcpyHostToDevice);
cudaMemcpyAsync(cuda_q_invperm, cpu_x_map_inv16, height * sizeof(uint16_t), cudaMemcpyHostToDevice);
// Rearrange rows in w
dim3 blockDim, gridDim;
blockDim.x = THREADS_X;
blockDim.y = 1;
gridDim.x = DIVIDE(width, THREADS_X);
gridDim.y = height / 8;
make_sequential_kernel<<<gridDim, blockDim>>>
(
cuda_q_weight,
cuda_new_qweight,
cuda_q_perm,
height / 8,
width
);
// Replace qweights
cudaMemcpyAsync(cuda_q_weight, cuda_new_qweight, height / 8 * width * sizeof(uint32_t), cudaMemcpyDeviceToDevice);
// Cleanup
cudaDeviceSynchronize();
cudaFree(cuda_new_qweight);
free(cpu_g_idx_map);
free(cpu_x_map);
free(cpu_x_map_inv);
return true;
}
| text-generation-inference/server/exllamav2_kernels/exllamav2_kernels/cuda/q_matrix.cu/0 | {
"file_path": "text-generation-inference/server/exllamav2_kernels/exllamav2_kernels/cuda/q_matrix.cu",
"repo_id": "text-generation-inference",
"token_count": 10391
} | 204 |
import torch
from loguru import logger
from transformers.configuration_utils import PretrainedConfig
from transformers.models.auto import modeling_auto
from typing import Optional
from text_generation_server.utils.speculate import get_speculate, set_speculate
from text_generation_server.models.model import Model
from text_generation_server.models.causal_lm import CausalLM
from text_generation_server.models.flash_causal_lm import FlashCausalLM
from text_generation_server.models.bloom import BLOOMSharded
from text_generation_server.models.mpt import MPTSharded
from text_generation_server.models.seq2seq_lm import Seq2SeqLM
from text_generation_server.models.rw import RW
from text_generation_server.models.opt import OPTSharded
from text_generation_server.models.galactica import GalacticaSharded
from text_generation_server.models.santacoder import SantaCoder
from text_generation_server.models.t5 import T5Sharded
from text_generation_server.models.gpt_neox import GPTNeoxSharded
from text_generation_server.models.phi import Phi
# The flag below controls whether to allow TF32 on matmul. This flag defaults to False
# in PyTorch 1.12 and later.
torch.backends.cuda.matmul.allow_tf32 = True
# The flag below controls whether to allow TF32 on cuDNN. This flag defaults to True.
torch.backends.cudnn.allow_tf32 = True
# Disable gradients
torch.set_grad_enabled(False)
__all__ = [
"Model",
"BLOOMSharded",
"CausalLM",
"FlashCausalLM",
"GalacticaSharded",
"Seq2SeqLM",
"SantaCoder",
"OPTSharded",
"T5Sharded",
"get_model",
]
FLASH_ATT_ERROR_MESSAGE = "{} requires Flash Attention enabled models."
FLASH_ATTENTION = True
try:
from text_generation_server.models.flash_rw import FlashRWSharded
from text_generation_server.models.flash_neox import FlashNeoXSharded
from text_generation_server.models.flash_llama import (
FlashLlama,
)
from text_generation_server.models.flash_santacoder import (
FlashSantacoderSharded,
)
from text_generation_server.models.idefics import IDEFICSSharded
from text_generation_server.models.flash_mistral import FlashMistral
from text_generation_server.models.flash_mixtral import FlashMixtral
from text_generation_server.models.flash_phi import FlashPhi
from text_generation_server.utils.flash_attn import HAS_FLASH_ATTN_V2_CUDA
except ImportError as e:
logger.warning(f"Could not import Flash Attention enabled models: {e}")
FLASH_ATTENTION = False
HAS_FLASH_ATTN_V2_CUDA = False
if FLASH_ATTENTION:
__all__.append(FlashNeoXSharded)
__all__.append(FlashRWSharded)
__all__.append(FlashSantacoderSharded)
__all__.append(FlashLlama)
__all__.append(IDEFICSSharded)
__all__.append(FlashMistral)
__all__.append(FlashMixtral)
__all__.append(FlashPhi)
def get_model(
model_id: str,
revision: Optional[str],
sharded: bool,
quantize: Optional[str],
speculate: Optional[int],
dtype: Optional[str],
trust_remote_code: bool,
) -> Model:
if dtype is None:
# Keep it as default for now and let
# every model resolve their own default dtype.
dtype = None
elif dtype == "float16":
dtype = torch.float16
elif dtype == "bfloat16":
dtype = torch.bfloat16
else:
raise RuntimeError(f"Unknown dtype {dtype}")
if speculate is not None:
set_speculate(speculate)
else:
set_speculate(0)
if "facebook/galactica" in model_id:
return GalacticaSharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
if model_id.startswith("bigcode/"):
if FLASH_ATTENTION:
return FlashSantacoderSharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
elif sharded:
raise NotImplementedError(
FLASH_ATT_ERROR_MESSAGE.format("Sharded Santacoder")
)
else:
return SantaCoder(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
config_dict, _ = PretrainedConfig.get_config_dict(
model_id, revision=revision, trust_remote_code=trust_remote_code
)
use_medusa = None
if "medusa_num_heads" in config_dict:
use_medusa = model_id
model_id = config_dict["base_model_name_or_path"]
revision = "main"
speculate_medusa = config_dict["medusa_num_heads"]
if speculate is not None:
if speculate > speculate_medusa:
raise RuntimeError(
"Speculate is set to `{speculate}` but this medusa models only has `{speculate_medusa}` heads, please make them match"
)
else:
set_speculate(speculate)
else:
set_speculate(speculate_medusa)
config_dict, _ = PretrainedConfig.get_config_dict(
model_id, revision=revision, trust_remote_code=trust_remote_code
)
method = "medusa"
else:
method = "n-gram"
speculate = get_speculate()
if speculate > 0:
logger.info(f"Using speculation {method} with {speculate} input ids.")
model_type = config_dict["model_type"]
if model_type == "gpt_bigcode":
if FLASH_ATTENTION:
return FlashSantacoderSharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
elif sharded:
raise NotImplementedError(
FLASH_ATT_ERROR_MESSAGE.format("Sharded Santacoder")
)
else:
return SantaCoder(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
if model_type == "bloom":
return BLOOMSharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
elif model_type == "mpt":
return MPTSharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
elif model_type == "gpt_neox":
if FLASH_ATTENTION:
return FlashNeoXSharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
elif sharded:
return GPTNeoxSharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
else:
return CausalLM(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
elif model_type == "phi":
if FLASH_ATTENTION:
return FlashPhi(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
use_medusa=use_medusa,
)
else:
return CausalLM(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
elif model_type == "phi-msft":
if FLASH_ATTENTION:
raise NotImplementedError(
"Legacy phi-msft is not supported with Flash Attention"
)
else:
return Phi(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
elif model_type == "llama" or model_type == "baichuan":
if FLASH_ATTENTION:
return FlashLlama(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
use_medusa=use_medusa,
)
elif sharded:
raise NotImplementedError(FLASH_ATT_ERROR_MESSAGE.format("Sharded Llama"))
else:
return CausalLM(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
if model_type in ["RefinedWeb", "RefinedWebModel", "falcon"]:
if sharded:
if FLASH_ATTENTION:
if config_dict.get("alibi", False):
raise NotImplementedError("sharded is not supported for this model")
return FlashRWSharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
raise NotImplementedError(FLASH_ATT_ERROR_MESSAGE.format(f"Sharded Falcon"))
else:
if FLASH_ATTENTION and not config_dict.get("alibi", False):
return FlashRWSharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
else:
return RW(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
if model_type == "mistral":
sliding_window = config_dict.get("sliding_window", -1)
if (
(sliding_window is None or sliding_window == -1) and FLASH_ATTENTION
) or HAS_FLASH_ATTN_V2_CUDA:
return FlashMistral(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
if model_type == "mixtral":
sliding_window = config_dict.get("sliding_window", -1)
if (
(sliding_window is None or sliding_window == -1) and FLASH_ATTENTION
) or HAS_FLASH_ATTN_V2_CUDA:
return FlashMixtral(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
if model_type == "opt":
return OPTSharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
if model_type == "t5":
return T5Sharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
if model_type == "idefics":
if FLASH_ATTENTION:
return IDEFICSSharded(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
else:
raise NotImplementedError(FLASH_ATT_ERROR_MESSAGE.format("Idefics"))
if sharded:
raise NotImplementedError("sharded is not supported for AutoModel")
if quantize == "gptq":
raise NotImplementedError(
"gptq quantization is not supported for AutoModel, you can try to quantize it with `text-generation-server quantize ORIGINAL_MODEL_ID NEW_MODEL_ID`"
)
if quantize == "awq":
raise NotImplementedError("awq quantization is not supported for AutoModel")
elif (quantize == "bitsandbytes-fp4") or (quantize == "bitsandbytes-nf4"):
raise NotImplementedError("4bit quantization is not supported for AutoModel")
elif quantize == "eetq":
raise NotImplementedError("Eetq quantization is not supported for AutoModel")
if model_type in modeling_auto.MODEL_FOR_CAUSAL_LM_MAPPING_NAMES:
return CausalLM(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
if model_type in modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES:
return Seq2SeqLM(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
auto_map = config_dict.get("auto_map", None)
if trust_remote_code and auto_map is not None:
if "AutoModelForCausalLM" in auto_map.keys():
return CausalLM(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
if "AutoModelForSeq2SeqLM" in auto_map.keys():
return Seq2SeqLM(
model_id,
revision,
quantize=quantize,
dtype=dtype,
trust_remote_code=trust_remote_code,
)
raise ValueError(f"Unsupported model type {model_type}")
| text-generation-inference/server/text_generation_server/models/__init__.py/0 | {
"file_path": "text-generation-inference/server/text_generation_server/models/__init__.py",
"repo_id": "text-generation-inference",
"token_count": 7094
} | 205 |
# This code was adapted from https://github.com/lucidrains/flamingo-pytorch licensed under the MIT License.
#
# MIT License
#
# Copyright (c) 2020 The Google AI Language Team Authors, The HuggingFace Inc. team and github/lonePatient
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
"""
Generic interface to various configurations of the Perceiver Resampler, that simply takes in a series of (potentially
time-indexed) contextual embeddings, and "resamples" (compresses) them down to a pre-specified number of latents! Note
that the Perceiver in general resamples based solely off the *long-range* context; there's a nice opportunity here to
prime the Perceiver Resampler with say a single layer's worth of language embeddings (the target domain), and use that
to softly "retrieve & compress" what we need --> this would be a novel contribution we should explore.
References:
- DeepMind's Flamingo: https://www.deepmind.com/blog/tackling-multiple-tasks-with-a-single-visual-language-model
- Code borrowed w/ love from: https://github.com/lucidrains/flamingo-pytorch
"""
from typing import Optional, Tuple
import torch
import torch.nn as nn
from text_generation_server.utils.layers import (
TensorParallelColumnLinear,
TensorParallelRowLinear,
)
EPS = 1e-5
class IdeficsPerceiverResampler(nn.Module):
def __init__(
self,
prefix,
config,
embed_dim: int,
depth: int,
n_heads: int,
head_dim: int,
n_latents: int,
weights,
) -> None:
"""
Instantiates a Perceiver Resampler that operates over a sequence of embeddings (say from a ResNet or ViT or
MAE) of a given dimension, performs `depth` blocks of cross-attention with a fixed `n_latents` inputs, then
returns a Tensor of shape [bsz, n_latents, embed_dim]. :param embed_dim: Dimensionality of embeddings being fed
to the Perceiver Resampler (also dimensionality of latent embeddings *returned* by the Perceiver Resampler.
Could be e.g., VIT embed_dim, ResNet pool dim, and so on.
Args:
config (`IdeficsConfig`): config object
embed_dim (`int`): The size of each embedding vector
depth (`int`): Depth of the Perceiver Resampler (Transformer w/ cross attention). Should be shallow (< 3).
n_heads (`int`): Number of heads in each Transformer block (for multi-headed self-attention).
head_dim (`int`): Dimensionality of each head projection in the Transformer block.
n_latents (`int`):
Number of latent embeddings to resample ("compress") the input sequence to (usually < 128).
"""
super().__init__()
self.embed_dim, self.n_heads, self.head_dim, self.n_latents = (
embed_dim,
n_heads,
head_dim,
n_latents,
)
self.qk_layer_norms = config.perceiver_config.qk_layer_norms_perceiver
# Create Latents for Perceiver
self.latents = nn.Parameter(weights.get_tensor(f"{prefix}.latents"))
self.intermediate_dim = (
self.embed_dim * 4
if not hasattr(config.vision_config, "embed_dim")
else config.vision_config.embed_dim * 4
)
# Create Transformer Blocks
self.blocks = nn.ModuleList(
[
nn.ModuleList(
[
IdeficsPerceiverAttention(
prefix=f"{prefix}.blocks.{layer_id}.0",
config=config,
embed_dim=self.embed_dim,
n_heads=self.n_heads,
head_dim=self.head_dim,
qk_layer_norms=self.qk_layer_norms,
weights=weights,
),
IdeficsMLP(
prefix=f"{prefix}.blocks.{layer_id}.1",
intermediate_size=self.intermediate_dim,
config=config,
weights=weights,
),
]
)
for layer_id in range(depth)
]
)
self.layer_norm = nn.LayerNorm.load(
prefix=f"{prefix}.layer_norm", weights=weights, eps=EPS
)
def forward(self, context: torch.Tensor) -> torch.Tensor:
"""Resample arbitrary length context & *compress* down to self.n_latents latent embeddings"""
# einsum.repeat(self.latents, "seq embed -> bsz seq embed", bsz=context.shape[0])
latents = self.latents.repeat(context.shape[0], 1, 1)
# Feed through Perceiver Attention blocks...
for attn, ff in self.blocks:
latents = attn(context, latents) + latents
latents = ff(latents) + latents
return self.layer_norm(latents)
class IdeficsPerceiverAttention(nn.Module):
def __init__(
self,
prefix,
config,
embed_dim: int,
n_heads: int,
head_dim: int,
qk_layer_norms: bool,
weights,
) -> None:
"""Perceiver Cross-Attention Module --> let long-form inputs be `context`, resampled embeddings be `latents`"""
super().__init__()
self.embed_dim, self.n_heads, self.head_dim = embed_dim, n_heads, head_dim
self.qk_layer_norms = qk_layer_norms
# Normalization & Scaling
self.context_layer_norm = nn.LayerNorm.load(
prefix=f"{prefix}.context_layer_norm", weights=weights, eps=EPS
)
self.latents_layer_norm = nn.LayerNorm.load(
prefix=f"{prefix}.latents_layer_norm", weights=weights, eps=EPS
)
if self.qk_layer_norms:
self.q_layer_norm = nn.LayerNorm.load(
prefix=f"{prefix}.q_layer_norm", weights=weights, eps=EPS
)
self.k_layer_norm = nn.LayerNorm.load(
prefix=f"{prefix}.k_layer_norm", weights=weights, eps=EPS
)
self.qk_scale = self.head_dim**-0.5
process_group = weights.process_group
if n_heads % weights.process_group.size() != 0:
raise ValueError(
f"`num_heads` must be divisible by `num_shards` (got `num_heads`: {n_heads} "
f"and `num_shards`: {weights.process_group.size()}"
)
self.n_heads //= weights.process_group.size()
# Q, K, V Projection (no bias -- detail from Perceiver/Flamingo Papers).
self.q_proj = TensorParallelColumnLinear.load(
config=config, prefix=f"{prefix}.q_proj", weights=weights, bias=False
)
self.k_proj = TensorParallelColumnLinear.load(
config=config, prefix=f"{prefix}.k_proj", weights=weights, bias=False
)
self.v_proj = TensorParallelColumnLinear.load(
config=config, prefix=f"{prefix}.v_proj", weights=weights, bias=False
)
self.output_proj = TensorParallelRowLinear.load(
config=config, prefix=f"{prefix}.output_proj", weights=weights, bias=False
)
def forward(self, context: torch.Tensor, latents: torch.Tensor) -> torch.Tensor:
"""
Runs Perceiver Self-Attention, with special (context, latents) appended along the `seq` dimension!
Args:
context (`torch.Tensor`):
Tensor of shape `[bsz, seq, embed_dim]` representing long-form context to resample.
latents (`torch.Tensor`):
Tensor of shape `[bsz, n_latents, embed_dim]` representing fixed length latents to compress to.
Returns:
`torch.Tensor`: Tensor of shape `[bsz, n_latents, embed_dim]` representing attention over latents w/ cross
from context.
"""
context = self.context_layer_norm(context)
latents = self.latents_layer_norm(latents)
batch_size, seq_length, embed_dim = context.shape[:3]
# Query, Key, Value Projections --> Note that in Flamingo, latents are *concatenated* with context prior to attn!
# Note: This results in queries w/ `seq = n_latents`, and keys, values with `seq = len(context) + n_latents`
q = self.q_proj(latents)
k = self.k_proj(torch.cat([context, latents], dim=-2))
v = self.v_proj(torch.cat([context, latents], dim=-2))
# Multiheaded Self-Attention w/ stable softmax (subtract per-row max -- `amax` -- before softmax call)
# =>> `attn` should be a 2D matrix of shape [n_latents x (context + n_latents)]
# einsum.rearrange(x, "bsz seq (heads embed) -> bsz heads seq embed", heads=self.n_heads)
q, k, v = [
x.reshape(batch_size, x.shape[1], self.n_heads, self.head_dim).transpose(
1, 2
)
for x in (q, k, v)
]
if self.qk_layer_norms:
q = self.q_layer_norm(q)
k = self.k_layer_norm(k)
scores = torch.einsum("... i d, ... j d -> ... i j", q * self.qk_scale, k)
stabilized_scores = scores - (scores.amax(dim=-1, keepdim=True).detach())
attn = stabilized_scores.softmax(dim=-1)
# Attend & project back to output...
resampled = torch.einsum("... i j, ... j d -> ... i d", attn, v)
# einsum.rearrange(resampled, "bsz heads seq embed -> bsz seq (heads embed)", heads=self.n_heads)
return self.output_proj(resampled.transpose(1, 2).flatten(-2))
class IdeficsMLP(nn.Module):
def __init__(
self,
prefix,
intermediate_size,
config,
weights,
):
"""Simple MLP block with intermediate_size and embedding size"""
super().__init__()
self.embed_dim = config.vision_config.embed_dim
self.ln = nn.LayerNorm.load(prefix=f"{prefix}.ln", weights=weights, eps=EPS)
self.fc = TensorParallelColumnLinear.load(
config=config,
prefix=f"{prefix}.fc",
weights=weights,
bias=False,
)
self.act = nn.ReLU()
self.c_proj = TensorParallelRowLinear.load(
config=config,
prefix=f"{prefix}.c_proj",
weights=weights,
bias=False,
)
def forward(
self, hidden_states: Optional[Tuple[torch.FloatTensor]]
) -> torch.FloatTensor:
hidden_states = self.ln(hidden_states)
hidden_states = self.fc(hidden_states)
hidden_states = self.act(hidden_states)
hidden_states = self.c_proj(hidden_states)
return hidden_states
| text-generation-inference/server/text_generation_server/models/custom_modeling/idefics_perceiver.py/0 | {
"file_path": "text-generation-inference/server/text_generation_server/models/custom_modeling/idefics_perceiver.py",
"repo_id": "text-generation-inference",
"token_count": 5171
} | 206 |
import re
import torch
import torch.distributed
from typing import List, Optional, Type
from transformers import (
AutoTokenizer,
AutoConfig,
PreTrainedTokenizerBase,
)
from text_generation_server.models import CausalLM
from text_generation_server.models.causal_lm import CausalLMBatch
from text_generation_server.pb import generate_pb2
from text_generation_server.models.custom_modeling.opt_modeling import OPTForCausalLM
from text_generation_server.utils import (
NextTokenChooser,
StoppingCriteria,
initialize_torch_distributed,
weight_files,
Weights,
)
# CREDIT: Papers with code => https://github.com/paperswithcode/galai/blob/main/galai/utils.py
# we split individual characters inside special tokens like [START_DNA]
CUSTOM_SEQ_RE = re.compile(r"(\[START_(DNA|SMILES|I_SMILES|AMINO)])(.*?)(\[END_\2])")
# token added to implement a custom sequence tokenization. This token is added at
# corpus cleaning step and removed in pretokenization. The digits are added to increase the chance
# that they do not occur in the corpus. The digits are escaped so that the token does not appear
# literally in the source code in case we ever include it in the training data.
SPLIT_MARKER = f"SPL{1}T-TH{1}S-Pl3A5E"
def _insert_split_marker(m: re.Match):
"""
Applies split marker based on a regex match of special tokens such as
[START_DNA].
Parameters
----------
n : str
Input text to split
Returns
----------
str - the text with the split token added
"""
start_token, _, sequence, end_token = m.groups()
sequence = re.sub(r"(.)", rf"{SPLIT_MARKER}\1", sequence, flags=re.DOTALL)
return f"{start_token}{sequence}{SPLIT_MARKER}{end_token}"
def escape_custom_split_sequence(text):
"""
Applies custom splitting to the text for GALILEO's tokenization
Parameters
----------
text : str
Input text to split
Returns
----------
str - the text with the split token added
"""
return CUSTOM_SEQ_RE.sub(_insert_split_marker, text)
# END CREDIT
class GalacticaCausalLMBatch(CausalLMBatch):
@classmethod
def from_pb(
cls,
pb: generate_pb2.Batch,
tokenizer: PreTrainedTokenizerBase,
dtype: torch.dtype,
device: torch.device,
) -> "GalacticaCausalLMBatch":
inputs = []
next_token_choosers = []
stopping_criterias = []
prefix_offsets = []
top_n_tokens = []
read_offsets = []
requests_idx_mapping = {}
# Parse batch
max_truncation = 0
padding_right_offset = 0
max_decode_tokens = 0
for i, r in enumerate(pb.requests):
requests_idx_mapping[r.id] = i
# Add escape_custom_split_sequence to the CausalLMBatch logic
inputs.append(escape_custom_split_sequence(r.inputs))
next_token_choosers.append(NextTokenChooser.from_pb(r.parameters, device))
stopping_criteria = StoppingCriteria.from_pb(
r.stopping_parameters, tokenizer
)
stopping_criterias.append(stopping_criteria)
top_n_tokens.append(r.top_n_tokens)
max_truncation = max(max_truncation, r.truncate)
max_decode_tokens += stopping_criteria.max_new_tokens
padding_right_offset = max(
padding_right_offset, stopping_criteria.max_new_tokens
)
tokenized_inputs = tokenizer(
inputs,
return_tensors="pt",
padding=True,
return_token_type_ids=False,
truncation=True,
max_length=max_truncation,
).to(device)
for _ in pb.requests:
input_len = tokenized_inputs["input_ids"].shape[1]
prefix_offsets.append(0)
read_offsets.append(input_len)
input_lengths = tokenized_inputs["attention_mask"].sum(1)
max_input_length = input_lengths.max()
input_ids = tokenized_inputs["input_ids"]
# Allocate maximum attention_mask
attention_mask = input_ids.new_zeros(
(pb.size, max_input_length + padding_right_offset)
)
# Copy tokenizer attention_mask into fully allocated attention_mask
attention_mask[:, :max_input_length] = tokenized_inputs["attention_mask"]
position_ids = tokenized_inputs["attention_mask"].long().cumsum(-1) - 1
position_ids.masked_fill_(tokenized_inputs["attention_mask"] == 0, 1)
all_input_ids = tokenized_inputs["input_ids"].T.split(1, dim=1)
top_n_tokens_tensor = torch.tensor(
top_n_tokens, device=device, dtype=torch.int64
)
max_tokens = len(inputs) * max_input_length + max_decode_tokens
return cls(
batch_id=pb.id,
requests=pb.requests,
requests_idx_mapping=requests_idx_mapping,
input_ids=input_ids,
attention_mask=attention_mask,
position_ids=position_ids,
past_key_values=None,
all_input_ids=list(all_input_ids),
input_lengths=input_lengths.tolist(),
prefix_offsets=prefix_offsets,
read_offsets=read_offsets,
next_token_choosers=next_token_choosers,
stopping_criterias=stopping_criterias,
top_n_tokens=top_n_tokens,
top_n_tokens_tensor=top_n_tokens_tensor,
max_input_length=max_input_length.item(),
padding_right_offset=padding_right_offset,
max_tokens=max_tokens,
)
class GalacticaSharded(CausalLM):
def __init__(
self,
model_id: str,
revision: Optional[str] = None,
quantize: Optional[str] = None,
dtype: Optional[torch.dtype] = None,
trust_remote_code: bool = False,
):
self.process_group, rank, world_size = initialize_torch_distributed()
if torch.cuda.is_available():
device = torch.device(f"cuda:{rank}")
dtype = torch.float16 if dtype is None else dtype
else:
device = torch.device("cpu")
dtype = torch.float32 if dtype is None else dtype
tokenizer = AutoTokenizer.from_pretrained(
model_id,
revision=revision,
padding_side="left",
truncation_side="left",
trust_remote_code=trust_remote_code,
)
config = AutoConfig.from_pretrained(
model_id,
revision=revision,
tp_parallel=True,
trust_remote_code=trust_remote_code,
)
config.quantize = quantize
tokenizer.pad_token_id = config.pad_token_id
torch.distributed.barrier(group=self.process_group)
filenames = weight_files(model_id, revision=revision, extension=".safetensors")
weights = Weights(
filenames, device=device, dtype=dtype, process_group=self.process_group
)
if config.quantize == "gptq":
weights._set_gptq_params(model_id, revision)
model = OPTForCausalLM(config, weights)
torch.distributed.barrier(group=self.process_group)
super(CausalLM, self).__init__(
model=model,
tokenizer=tokenizer,
requires_padding=True,
dtype=dtype,
device=device,
rank=rank,
world_size=world_size,
)
@property
def batch_type(self) -> Type[CausalLMBatch]:
return GalacticaCausalLMBatch
def decode(self, generated_ids: List[int]) -> str:
# Do not skip special tokens as they are used for custom parsing rules of the generated text
return self.tokenizer.decode(
generated_ids, skip_special_tokens=False, clean_up_tokenization_spaces=False
)
def forward(
self, input_ids, attention_mask, position_ids, past_key_values: Optional = None
):
outputs = self.model.forward(
input_ids=input_ids,
attention_mask=attention_mask,
past_key_values=past_key_values,
use_cache=True,
)
return outputs.logits, outputs.past_key_values
| text-generation-inference/server/text_generation_server/models/galactica.py/0 | {
"file_path": "text-generation-inference/server/text_generation_server/models/galactica.py",
"repo_id": "text-generation-inference",
"token_count": 3723
} | 207 |
from text_generation_server.utils.convert import convert_file, convert_files
from text_generation_server.utils.dist import initialize_torch_distributed
from text_generation_server.utils.weights import Weights
from text_generation_server.utils.peft import download_and_unload_peft
from text_generation_server.utils.hub import (
weight_files,
weight_hub_files,
download_weights,
EntryNotFoundError,
LocalEntryNotFoundError,
RevisionNotFoundError,
)
from text_generation_server.utils.tokens import (
NextTokenChooser,
HeterogeneousNextTokenChooser,
StoppingCriteria,
StopSequenceCriteria,
FinishReason,
Sampling,
Greedy,
)
__all__ = [
"convert_file",
"convert_files",
"initialize_torch_distributed",
"weight_files",
"weight_hub_files",
"download_weights",
"download_and_unload_peft",
"EntryNotFoundError",
"HeterogeneousNextTokenChooser",
"LocalEntryNotFoundError",
"RevisionNotFoundError",
"Greedy",
"NextTokenChooser",
"Sampling",
"StoppingCriteria",
"StopSequenceCriteria",
"FinishReason",
"Weights",
]
| text-generation-inference/server/text_generation_server/utils/__init__.py/0 | {
"file_path": "text-generation-inference/server/text_generation_server/utils/__init__.py",
"repo_id": "text-generation-inference",
"token_count": 417
} | 208 |
import torch
# vllm imports
from vllm import cache_ops
from vllm import attention_ops
_PARTITION_SIZE = 512
def reshape_and_cache(
key: torch.Tensor,
value: torch.Tensor,
key_cache: torch.Tensor,
value_cache: torch.Tensor,
slots: torch.Tensor,
):
cache_ops.reshape_and_cache(key, value, key_cache, value_cache, slots)
def attention(
out: torch.Tensor,
query: torch.Tensor,
key_cache: torch.Tensor,
value_cache: torch.Tensor,
kv_head_mapping: torch.Tensor,
softmax_scale: float,
block_tables: torch.Tensor,
input_lengths: torch.Tensor,
max_s: int,
):
# Adapted from: https://github.com/vllm-project/vllm/blob/f8a1e39fae05ca610be8d5a78be9d40f5274e5fc/vllm/model_executor/layers/attention.py
# Copyright 2023 The vLLM team. All rights
# reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
# value_cache => [num_blocks, num_heads, head_size, block_size]
block_size = value_cache.shape[3]
num_seqs, num_heads, head_size = query.shape
max_num_partitions = (max_s + _PARTITION_SIZE - 1) // _PARTITION_SIZE
# NOTE(woosuk): We use a simple heuristic to decide whether to use
# PagedAttention V1 or V2. If the number of partitions is 1, we use
# V1 to avoid the overhead of reduction. Also, if the number of
# sequences or heads is large, we use V1 since there is enough work
# to parallelize.
use_v1 = max_num_partitions == 1 or num_seqs * num_heads > 512
if use_v1:
attention_ops.paged_attention_v1(
out,
query,
key_cache,
value_cache,
kv_head_mapping,
softmax_scale,
block_tables,
input_lengths,
block_size,
max_s,
None,
)
else:
# Run PagedAttention V2.
assert _PARTITION_SIZE % block_size == 0
tmp_output = torch.empty(
size=(num_seqs, num_heads, max_num_partitions, head_size),
dtype=out.dtype,
device=out.device,
)
exp_sums = torch.empty(
size=(num_seqs, num_heads, max_num_partitions),
dtype=torch.float32,
device=out.device,
)
max_logits = torch.empty_like(exp_sums)
attention_ops.paged_attention_v2(
out,
exp_sums,
max_logits,
tmp_output,
query,
key_cache,
value_cache,
kv_head_mapping,
softmax_scale,
block_tables,
input_lengths,
block_size,
max_s,
None,
)
| text-generation-inference/server/text_generation_server/utils/paged_attention.py/0 | {
"file_path": "text-generation-inference/server/text_generation_server/utils/paged_attention.py",
"repo_id": "text-generation-inference",
"token_count": 1485
} | 209 |
[package]
authors = ["Nicolas Patry <[email protected]>"]
edition = "2021"
name = "node"
version = "0.15.2-dev.0"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[lib]
crate-type = ["cdylib"]
[dependencies]
napi = "2"
napi-derive = "2"
serde = { version = "1.0.163", features = ["derive"] }
tokenizers = { path = "../../tokenizers/" }
[build-dependencies]
napi-build = "2"
[profile.release]
lto = true
| tokenizers/bindings/node/Cargo.toml/0 | {
"file_path": "tokenizers/bindings/node/Cargo.toml",
"repo_id": "tokenizers",
"token_count": 200
} | 210 |
import { prependNormalizer, stripAccentsNormalizer, stripNormalizer } from '../../'
describe('stripNormalizer', () => {
it('instantiates with no parameters', () => {
const normalizer = stripNormalizer()
expect(normalizer.constructor.name).toEqual('Normalizer')
})
it('accepts `undefined` as first parameter', () => {
expect(stripNormalizer(undefined)).toBeDefined()
})
it('accepts `undefined` as second parameter', () => {
expect(stripNormalizer(false, undefined)).toBeDefined()
})
it('instantiates with one parameter', () => {
const normalizer = stripNormalizer(false)
expect(normalizer.constructor.name).toEqual('Normalizer')
})
it('instantiates with two parameters', () => {
const normalizer = stripNormalizer(false, true)
expect(normalizer.constructor.name).toEqual('Normalizer')
})
it('prepend instantiates with one parameter', () => {
const normalizer = prependNormalizer('_')
expect(normalizer.constructor.name).toEqual('Normalizer')
expect(normalizer.normalizeString('Hello')).toEqual('_Hello')
})
it('can normalize strings', () => {
const normalizer = stripNormalizer()
expect(normalizer.normalizeString(' Hello there ')).toEqual('Hello there')
})
})
describe('stripAccentsNormalizer', () => {
it('initialize', () => {
const normalizer = stripAccentsNormalizer()
expect(normalizer.constructor.name).toEqual('Normalizer')
})
})
| tokenizers/bindings/node/lib/bindings/normalizers.test.ts/0 | {
"file_path": "tokenizers/bindings/node/lib/bindings/normalizers.test.ts",
"repo_id": "tokenizers",
"token_count": 468
} | 211 |
{
"name": "tokenizers-linux-arm-gnueabihf",
"version": "0.13.4-rc1",
"os": [
"linux"
],
"cpu": [
"arm"
],
"main": "tokenizers.linux-arm-gnueabihf.node",
"files": [
"tokenizers.linux-arm-gnueabihf.node"
],
"description": "Tokenizers platform specific bindings",
"keywords": [
"napi-rs",
"NAPI",
"N-API",
"Rust",
"node-addon",
"node-addon-api"
],
"license": "MIT",
"engines": {
"node": ">= 10"
},
"publishConfig": {
"registry": "https://registry.npmjs.org/",
"access": "public"
},
"repository": "tokenizers"
} | tokenizers/bindings/node/npm/linux-arm-gnueabihf/package.json/0 | {
"file_path": "tokenizers/bindings/node/npm/linux-arm-gnueabihf/package.json",
"repo_id": "tokenizers",
"token_count": 278
} | 212 |
tab_spaces = 2
| tokenizers/bindings/node/rustfmt.toml/0 | {
"file_path": "tokenizers/bindings/node/rustfmt.toml",
"repo_id": "tokenizers",
"token_count": 7
} | 213 |
export type TextInputSequence = string
export type PreTokenizedInputSequence = string[]
export type InputSequence = TextInputSequence | PreTokenizedInputSequence
export type TextEncodeInput = TextInputSequence | [TextInputSequence, TextInputSequence]
export type PreTokenizedEncodeInput = PreTokenizedInputSequence | [PreTokenizedInputSequence, PreTokenizedInputSequence]
export type EncodeInput = TextEncodeInput | PreTokenizedEncodeInput
| tokenizers/bindings/node/types.ts/0 | {
"file_path": "tokenizers/bindings/node/types.ts",
"repo_id": "tokenizers",
"token_count": 114
} | 214 |
from enum import Enum
from typing import List, Tuple, Union
Offsets = Tuple[int, int]
TextInputSequence = str
"""A :obj:`str` that represents an input sequence """
PreTokenizedInputSequence = Union[List[str], Tuple[str]]
"""A pre-tokenized input sequence. Can be one of:
- A :obj:`List` of :obj:`str`
- A :obj:`Tuple` of :obj:`str`
"""
TextEncodeInput = Union[
TextInputSequence,
Tuple[TextInputSequence, TextInputSequence],
List[TextInputSequence],
]
"""Represents a textual input for encoding. Can be either:
- A single sequence: :data:`~tokenizers.TextInputSequence`
- A pair of sequences:
- A :obj:`Tuple` of :data:`~tokenizers.TextInputSequence`
- Or a :obj:`List` of :data:`~tokenizers.TextInputSequence` of size 2
"""
PreTokenizedEncodeInput = Union[
PreTokenizedInputSequence,
Tuple[PreTokenizedInputSequence, PreTokenizedInputSequence],
List[PreTokenizedInputSequence],
]
"""Represents a pre-tokenized input for encoding. Can be either:
- A single sequence: :data:`~tokenizers.PreTokenizedInputSequence`
- A pair of sequences:
- A :obj:`Tuple` of :data:`~tokenizers.PreTokenizedInputSequence`
- Or a :obj:`List` of :data:`~tokenizers.PreTokenizedInputSequence` of size 2
"""
InputSequence = Union[TextInputSequence, PreTokenizedInputSequence]
"""Represents all the possible types of input sequences for encoding. Can be:
- When ``is_pretokenized=False``: :data:`~TextInputSequence`
- When ``is_pretokenized=True``: :data:`~PreTokenizedInputSequence`
"""
EncodeInput = Union[TextEncodeInput, PreTokenizedEncodeInput]
"""Represents all the possible types of input for encoding. Can be:
- When ``is_pretokenized=False``: :data:`~TextEncodeInput`
- When ``is_pretokenized=True``: :data:`~PreTokenizedEncodeInput`
"""
class OffsetReferential(Enum):
ORIGINAL = "original"
NORMALIZED = "normalized"
class OffsetType(Enum):
BYTE = "byte"
CHAR = "char"
class SplitDelimiterBehavior(Enum):
REMOVED = "removed"
ISOLATED = "isolated"
MERGED_WITH_PREVIOUS = "merged_with_previous"
MERGED_WITH_NEXT = "merged_with_next"
CONTIGUOUS = "contiguous"
from .tokenizers import (
AddedToken,
Encoding,
NormalizedString,
PreTokenizedString,
Regex,
Token,
Tokenizer,
decoders,
models,
normalizers,
pre_tokenizers,
processors,
trainers,
__version__,
)
from .implementations import (
BertWordPieceTokenizer,
ByteLevelBPETokenizer,
CharBPETokenizer,
SentencePieceBPETokenizer,
SentencePieceUnigramTokenizer,
)
| tokenizers/bindings/python/py_src/tokenizers/__init__.py/0 | {
"file_path": "tokenizers/bindings/python/py_src/tokenizers/__init__.py",
"repo_id": "tokenizers",
"token_count": 984
} | 215 |
# Generated content DO NOT EDIT
class PreTokenizer:
"""
Base class for all pre-tokenizers
This class is not supposed to be instantiated directly. Instead, any implementation of a
PreTokenizer will return an instance of this class when instantiated.
"""
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
class BertPreTokenizer(PreTokenizer):
"""
BertPreTokenizer
This pre-tokenizer splits tokens on spaces, and also on punctuation.
Each occurence of a punctuation character will be treated separately.
"""
def __init__(self):
pass
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
class ByteLevel(PreTokenizer):
"""
ByteLevel PreTokenizer
This pre-tokenizer takes care of replacing all bytes of the given string
with a corresponding representation, as well as splitting into words.
Args:
add_prefix_space (:obj:`bool`, `optional`, defaults to :obj:`True`):
Whether to add a space to the first word if there isn't already one. This
lets us treat `hello` exactly like `say hello`.
use_regex (:obj:`bool`, `optional`, defaults to :obj:`True`):
Set this to :obj:`False` to prevent this `pre_tokenizer` from using
the GPT2 specific regexp for spliting on whitespace.
"""
def __init__(self, add_prefix_space=True, use_regex=True):
pass
@staticmethod
def alphabet():
"""
Returns the alphabet used by this PreTokenizer.
Since the ByteLevel works as its name suggests, at the byte level, it
encodes each byte value to a unique visible character. This means that there is a
total of 256 different characters composing this alphabet.
Returns:
:obj:`List[str]`: A list of characters that compose the alphabet
"""
pass
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
class CharDelimiterSplit(PreTokenizer):
"""
This pre-tokenizer simply splits on the provided char. Works like `.split(delimiter)`
Args:
delimiter: str:
The delimiter char that will be used to split input
"""
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
class Digits(PreTokenizer):
"""
This pre-tokenizer simply splits using the digits in separate tokens
Args:
individual_digits (:obj:`bool`, `optional`, defaults to :obj:`False`):
If set to True, digits will each be separated as follows::
"Call 123 please" -> "Call ", "1", "2", "3", " please"
If set to False, digits will grouped as follows::
"Call 123 please" -> "Call ", "123", " please"
"""
def __init__(self, individual_digits=False):
pass
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
class Metaspace(PreTokenizer):
"""
Metaspace pre-tokenizer
This pre-tokenizer replaces any whitespace by the provided replacement character.
It then tries to split on these spaces.
Args:
replacement (:obj:`str`, `optional`, defaults to :obj:`â`):
The replacement character. Must be exactly one character. By default we
use the `â` (U+2581) meta symbol (Same as in SentencePiece).
add_prefix_space (:obj:`bool`, `optional`, defaults to :obj:`True`):
Whether to add a space to the first word if there isn't already one. This
lets us treat `hello` exactly like `say hello`.
"""
def __init__(self, replacement="_", add_prefix_space=True):
pass
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
class Punctuation(PreTokenizer):
"""
This pre-tokenizer simply splits on punctuation as individual characters.
Args:
behavior (:class:`~tokenizers.SplitDelimiterBehavior`):
The behavior to use when splitting.
Choices: "removed", "isolated" (default), "merged_with_previous", "merged_with_next",
"contiguous"
"""
def __init__(self, behavior="isolated"):
pass
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
class Sequence(PreTokenizer):
"""
This pre-tokenizer composes other pre_tokenizers and applies them in sequence
"""
def __init__(self, pretokenizers):
pass
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
class Split(PreTokenizer):
"""
Split PreTokenizer
This versatile pre-tokenizer splits using the provided pattern and
according to the provided behavior. The pattern can be inverted by
making use of the invert flag.
Args:
pattern (:obj:`str` or :class:`~tokenizers.Regex`):
A pattern used to split the string. Usually a string or a a regex built with `tokenizers.Regex`
behavior (:class:`~tokenizers.SplitDelimiterBehavior`):
The behavior to use when splitting.
Choices: "removed", "isolated", "merged_with_previous", "merged_with_next",
"contiguous"
invert (:obj:`bool`, `optional`, defaults to :obj:`False`):
Whether to invert the pattern.
"""
def __init__(self, pattern, behavior, invert=False):
pass
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
class UnicodeScripts(PreTokenizer):
"""
This pre-tokenizer splits on characters that belong to different language family
It roughly follows https://github.com/google/sentencepiece/blob/master/data/Scripts.txt
Actually Hiragana and Katakana are fused with Han, and 0x30FC is Han too.
This mimicks SentencePiece Unigram implementation.
"""
def __init__(self):
pass
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
class Whitespace(PreTokenizer):
"""
This pre-tokenizer simply splits using the following regex: `\w+|[^\w\s]+`
"""
def __init__(self):
pass
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
class WhitespaceSplit(PreTokenizer):
"""
This pre-tokenizer simply splits on the whitespace. Works like `.split()`
"""
def __init__(self):
pass
def pre_tokenize(self, pretok):
"""
Pre-tokenize a :class:`~tokenizers.PyPreTokenizedString` in-place
This method allows to modify a :class:`~tokenizers.PreTokenizedString` to
keep track of the pre-tokenization, and leverage the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you just want to see the result of
the pre-tokenization of a raw string, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize_str`
Args:
pretok (:class:`~tokenizers.PreTokenizedString):
The pre-tokenized string on which to apply this
:class:`~tokenizers.pre_tokenizers.PreTokenizer`
"""
pass
def pre_tokenize_str(self, sequence):
"""
Pre tokenize the given string
This method provides a way to visualize the effect of a
:class:`~tokenizers.pre_tokenizers.PreTokenizer` but it does not keep track of the
alignment, nor does it provide all the capabilities of the
:class:`~tokenizers.PreTokenizedString`. If you need some of these, you can use
:meth:`~tokenizers.pre_tokenizers.PreTokenizer.pre_tokenize`
Args:
sequence (:obj:`str`):
A string to pre-tokeize
Returns:
:obj:`List[Tuple[str, Offsets]]`:
A list of tuple with the pre-tokenized parts and their offsets
"""
pass
| tokenizers/bindings/python/py_src/tokenizers/pre_tokenizers/__init__.pyi/0 | {
"file_path": "tokenizers/bindings/python/py_src/tokenizers/pre_tokenizers/__init__.pyi",
"repo_id": "tokenizers",
"token_count": 9461
} | 216 |
use pyo3::exceptions;
use pyo3::prelude::*;
use pyo3::type_object::PyTypeInfo;
use std::fmt::{Display, Formatter, Result as FmtResult};
use tokenizers::tokenizer::Result;
#[derive(Debug)]
pub struct PyError(pub String);
impl PyError {
#[allow(dead_code)]
pub fn from(s: &str) -> Self {
PyError(String::from(s))
}
pub fn into_pyerr<T: PyTypeInfo>(self) -> PyErr {
PyErr::new::<T, _>(format!("{}", self))
}
}
impl Display for PyError {
fn fmt(&self, fmt: &mut Formatter) -> FmtResult {
write!(fmt, "{}", self.0)
}
}
impl std::error::Error for PyError {}
pub struct ToPyResult<T>(pub Result<T>);
impl<T> From<ToPyResult<T>> for PyResult<T> {
fn from(v: ToPyResult<T>) -> Self {
v.0.map_err(|e| exceptions::PyException::new_err(format!("{}", e)))
}
}
impl<T> ToPyResult<T> {
pub fn into_py(self) -> PyResult<T> {
self.into()
}
}
pub(crate) fn deprecation_warning(py: Python<'_>, version: &str, message: &str) -> PyResult<()> {
let deprecation_warning = py.import("builtins")?.getattr("DeprecationWarning")?;
let full_message = format!("Deprecated in {}: {}", version, message);
pyo3::PyErr::warn(py, deprecation_warning, &full_message, 0)
}
| tokenizers/bindings/python/src/error.rs/0 | {
"file_path": "tokenizers/bindings/python/src/error.rs",
"repo_id": "tokenizers",
"token_count": 531
} | 217 |
import pytest
from tokenizers import BertWordPieceTokenizer
from ..utils import bert_files, data_dir, multiprocessing_with_parallelism
class TestBertWordPieceTokenizer:
def test_basic_encode(self, bert_files):
tokenizer = BertWordPieceTokenizer.from_file(bert_files["vocab"])
# Encode with special tokens by default
output = tokenizer.encode("My name is John", "pair")
assert output.ids == [101, 2026, 2171, 2003, 2198, 102, 3940, 102]
assert output.tokens == [
"[CLS]",
"my",
"name",
"is",
"john",
"[SEP]",
"pair",
"[SEP]",
]
assert output.offsets == [
(0, 0),
(0, 2),
(3, 7),
(8, 10),
(11, 15),
(0, 0),
(0, 4),
(0, 0),
]
assert output.type_ids == [0, 0, 0, 0, 0, 0, 1, 1]
# Can encode without the special tokens
output = tokenizer.encode("My name is John", "pair", add_special_tokens=False)
assert output.ids == [2026, 2171, 2003, 2198, 3940]
assert output.tokens == ["my", "name", "is", "john", "pair"]
assert output.offsets == [(0, 2), (3, 7), (8, 10), (11, 15), (0, 4)]
assert output.type_ids == [0, 0, 0, 0, 1]
def test_multiprocessing_with_parallelism(self, bert_files):
tokenizer = BertWordPieceTokenizer.from_file(bert_files["vocab"])
multiprocessing_with_parallelism(tokenizer, False)
multiprocessing_with_parallelism(tokenizer, True)
def test_train_from_iterator(self):
text = ["A first sentence", "Another sentence", "And a last one"]
tokenizer = BertWordPieceTokenizer()
tokenizer.train_from_iterator(text, show_progress=False)
output = tokenizer.encode("A sentence")
assert output.tokens == ["a", "sentence"]
| tokenizers/bindings/python/tests/implementations/test_bert_wordpiece.py/0 | {
"file_path": "tokenizers/bindings/python/tests/implementations/test_bert_wordpiece.py",
"repo_id": "tokenizers",
"token_count": 919
} | 218 |
# Post-processors
<tokenizerslangcontent>
<python>
## BertProcessing
[[autodoc]] tokenizers.processors.BertProcessing
## ByteLevel
[[autodoc]] tokenizers.processors.ByteLevel
## RobertaProcessing
[[autodoc]] tokenizers.processors.RobertaProcessing
## TemplateProcessing
[[autodoc]] tokenizers.processors.TemplateProcessing
</python>
<rust>
The Rust API Reference is available directly on the [Docs.rs](https://docs.rs/tokenizers/latest/tokenizers/) website.
</rust>
<node>
The node API has not been documented yet.
</node>
</tokenizerslangcontent> | tokenizers/docs/source-doc-builder/api/post-processors.mdx/0 | {
"file_path": "tokenizers/docs/source-doc-builder/api/post-processors.mdx",
"repo_id": "tokenizers",
"token_count": 174
} | 219 |
Crates.io
----------------------------------------------------------------------------------------------------
ð€ Tokenizers is available on `crates.io <https://crates.io/crates/tokenizers>`__.
You just need to add it to your :obj:`Cargo.toml`::
tokenizers = "0.10"
| tokenizers/docs/source/installation/rust.inc/0 | {
"file_path": "tokenizers/docs/source/installation/rust.inc",
"repo_id": "tokenizers",
"token_count": 74
} | 220 |
{
"name": "create-wasm-app",
"version": "0.1.0",
"description": "create an app to consume rust-generated wasm packages",
"main": "index.js",
"bin": {
"create-wasm-app": ".bin/create-wasm-app.js"
},
"scripts": {
"build": "webpack --config webpack.config.js",
"start": "NODE_OPTIONS=--openssl-legacy-provider webpack-dev-server"
},
"repository": {
"type": "git",
"url": "git+https://github.com/rustwasm/create-wasm-app.git"
},
"keywords": ["webassembly", "wasm", "rust", "webpack"],
"author": "Ashley Williams <[email protected]>",
"license": "(MIT OR Apache-2.0)",
"bugs": {
"url": "https://github.com/rustwasm/create-wasm-app/issues"
},
"homepage": "https://github.com/rustwasm/create-wasm-app#readme",
"devDependencies": {
"copy-webpack-plugin": "^11.0.0",
"webpack": "^5.75.0",
"webpack-cli": "^5.0.1",
"webpack-dev-server": "^4.10.0"
},
"dependencies": {
"unstable_wasm": "file:../pkg"
}
}
| tokenizers/tokenizers/examples/unstable_wasm/www/package.json/0 | {
"file_path": "tokenizers/tokenizers/examples/unstable_wasm/www/package.json",
"repo_id": "tokenizers",
"token_count": 516
} | 221 |
#![allow(clippy::map_entry)]
use super::{Pair, WithFirstLastIterator, Word, BPE};
use crate::parallelism::*;
use crate::tokenizer::{AddedToken, Result, Trainer};
use crate::utils::progress::{ProgressBar, ProgressStyle};
use serde::{Deserialize, Serialize};
use std::cmp::Ordering;
use std::collections::{BinaryHeap, HashMap, HashSet};
#[derive(Debug, Eq)]
struct Merge {
pair: Pair,
count: u32,
pos: HashSet<usize>,
}
impl PartialEq for Merge {
fn eq(&self, other: &Self) -> bool {
self.count == other.count && self.pair == other.pair
}
}
impl PartialOrd for Merge {
fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
Some(self.cmp(other))
}
}
impl Ord for Merge {
fn cmp(&self, other: &Self) -> Ordering {
if self.count != other.count {
self.count.cmp(&other.count)
} else {
// Here we want ascending order
other.pair.cmp(&self.pair)
}
}
}
struct Config {
min_frequency: u32,
vocab_size: usize,
show_progress: bool,
special_tokens: Vec<AddedToken>,
limit_alphabet: Option<usize>,
initial_alphabet: HashSet<char>,
continuing_subword_prefix: Option<String>,
end_of_word_suffix: Option<String>,
max_token_length: Option<usize>,
}
/// A `BpeTrainerBuilder` can be used to create a `BpeTrainer` with a custom
/// configuration.
pub struct BpeTrainerBuilder {
config: Config,
}
impl Default for BpeTrainerBuilder {
fn default() -> Self {
Self {
config: Config {
min_frequency: 0,
vocab_size: 30000,
show_progress: true,
special_tokens: vec![],
limit_alphabet: None,
initial_alphabet: HashSet::new(),
continuing_subword_prefix: None,
end_of_word_suffix: None,
max_token_length: None,
},
}
}
}
impl BpeTrainerBuilder {
/// Constructs a new `BpeTrainerBuilder`
pub fn new() -> Self {
Self::default()
}
/// Set the expected minimum frequency
#[must_use]
pub fn min_frequency(mut self, frequency: u32) -> Self {
self.config.min_frequency = frequency;
self
}
/// Set the vocabulary size
#[must_use]
pub fn vocab_size(mut self, size: usize) -> Self {
self.config.vocab_size = size;
self
}
/// Set whether to show progress
#[must_use]
pub fn show_progress(mut self, show: bool) -> Self {
self.config.show_progress = show;
self
}
/// Set the special tokens
#[must_use]
pub fn special_tokens(mut self, tokens: Vec<AddedToken>) -> Self {
self.config.special_tokens = tokens;
self
}
/// Set whether to limit the alphabet
#[must_use]
pub fn limit_alphabet(mut self, limit: usize) -> Self {
self.config.limit_alphabet = Some(limit);
self
}
/// Set the initial alphabet
#[must_use]
pub fn initial_alphabet(mut self, alphabet: HashSet<char>) -> Self {
self.config.initial_alphabet = alphabet;
self
}
/// Set the continuing_subword_prefix
#[must_use]
pub fn continuing_subword_prefix(mut self, prefix: String) -> Self {
self.config.continuing_subword_prefix = Some(prefix);
self
}
/// Set the end_of_word_suffix
#[must_use]
pub fn end_of_word_suffix(mut self, suffix: String) -> Self {
self.config.end_of_word_suffix = Some(suffix);
self
}
/// Set max_token_length
#[must_use]
pub fn max_token_length(mut self, max_token_length: Option<usize>) -> Self {
self.config.max_token_length = max_token_length;
self
}
/// Constructs the final BpeTrainer
pub fn build(self) -> BpeTrainer {
BpeTrainer {
min_frequency: self.config.min_frequency,
vocab_size: self.config.vocab_size,
show_progress: self.config.show_progress,
special_tokens: self.config.special_tokens,
limit_alphabet: self.config.limit_alphabet,
initial_alphabet: self.config.initial_alphabet,
continuing_subword_prefix: self.config.continuing_subword_prefix,
end_of_word_suffix: self.config.end_of_word_suffix,
max_token_length: self.config.max_token_length,
words: HashMap::new(),
}
}
}
/// In charge of training a `BPE` model
///
/// # Examples
///
/// ```
/// use tokenizers::tokenizer::Trainer;
/// use tokenizers::models::bpe::{BPE, BpeTrainer};
///
/// let sequences = vec![ "Hello", "World" ];
///
/// let mut trainer = BpeTrainer::default();
/// trainer.feed(sequences.iter(), |s| Ok(vec![s.to_owned()]));
///
/// let mut model = BPE::default();
/// let special_tokens = trainer.train(&mut model).unwrap();
/// ```
#[non_exhaustive]
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize, Eq)]
pub struct BpeTrainer {
/// The minimum frequency a pair must have to produce a merge operation
pub min_frequency: u32,
/// The target vocabulary size
pub vocab_size: usize,
/// Whether to show progress while training
pub show_progress: bool,
/// A list of special tokens that the model should know of
pub special_tokens: Vec<AddedToken>,
/// Whether to limit the number of initial tokens that can be kept before computing merges
pub limit_alphabet: Option<usize>,
/// The initial alphabet we want absolutely to include. This allows to cover
/// some characters that are not necessarily in the training set
pub initial_alphabet: HashSet<char>,
/// An optional prefix to use on any subword that exist only behind another one
pub continuing_subword_prefix: Option<String>,
/// An optional suffix to caracterize and end-of-word subword
pub end_of_word_suffix: Option<String>,
/// An optional parameter to limit the max length of any single token
pub max_token_length: Option<usize>,
words: HashMap<String, u32>,
}
impl Default for BpeTrainer {
fn default() -> Self {
Self::builder().build()
}
}
impl BpeTrainer {
pub fn new(min_frequency: u32, vocab_size: usize) -> Self {
Self {
min_frequency,
vocab_size,
..Default::default()
}
}
pub fn builder() -> BpeTrainerBuilder {
BpeTrainerBuilder::new()
}
/// Setup a progress bar if asked to show progress
fn setup_progress(&self) -> Option<ProgressBar> {
if self.show_progress {
let p = ProgressBar::new(0);
p.set_style(
ProgressStyle::default_bar()
.template("[{elapsed_precise}] {msg:<30!} {wide_bar} {pos:<9!}/{len:>9!}")
.expect("Invalid progress template"),
);
Some(p)
} else {
None
}
}
/// Set the progress bar in the finish state
fn finalize_progress(&self, p: &Option<ProgressBar>, final_len: usize) {
if let Some(p) = p {
p.set_length(final_len as u64);
p.finish();
println!();
}
}
/// Update the progress bar with the new provided length and message
fn update_progress(&self, p: &Option<ProgressBar>, len: usize, message: &'static str) {
if let Some(p) = p {
p.set_message(message);
p.set_length(len as u64);
p.reset();
}
}
/// Add the provided special tokens to the initial vocabulary
fn add_special_tokens(&self, w2id: &mut HashMap<String, u32>, id2w: &mut Vec<String>) {
for token in &self.special_tokens {
if !w2id.contains_key(&token.content) {
id2w.push(token.content.to_owned());
w2id.insert(token.content.to_owned(), (id2w.len() - 1) as u32);
}
}
}
/// Compute the initial alphabet and limit it if relevant
fn compute_alphabet(
&self,
wc: &HashMap<String, u32>,
w2id: &mut HashMap<String, u32>,
id2w: &mut Vec<String>,
) {
// Compute the alphabet from seen words
let mut alphabet: HashMap<char, usize> = HashMap::new();
for (word, count) in wc {
for c in word.chars() {
alphabet
.entry(c)
.and_modify(|cnt| *cnt += *count as usize)
.or_insert(*count as usize);
}
}
// Also include anything from the provided initial alphabet
for c in &self.initial_alphabet {
alphabet
.entry(*c)
.and_modify(|cnt| *cnt = std::usize::MAX)
.or_insert(std::usize::MAX);
}
let mut kept = alphabet.iter().collect::<Vec<_>>();
// Compute the number of chars to remove from the alphabet
// If `limit_alphabet < initial_alphabet.len()`, some of these initial characters
// will be removed
let to_remove = self
.limit_alphabet
.map(|limit| {
if alphabet.len() > limit {
alphabet.len() - limit
} else {
0
}
})
.unwrap_or(0);
// Remove the unwanted chars
if to_remove > 0 {
kept.sort_unstable_by_key(|k| *k.1);
kept.drain(..to_remove);
}
// Keep the initial alphabet (sorted for determinism)
kept.sort_unstable_by_key(|k| (*k.0) as u32);
kept.into_iter().for_each(|(c, _)| {
let s = c.to_string();
if !w2id.contains_key(&s) {
id2w.push(s.clone());
w2id.insert(s, (id2w.len() - 1) as u32);
}
});
}
/// Tokenize words and add subwords to the vocabulary when relevant
fn tokenize_words(
&self,
wc: &HashMap<String, u32>,
w2id: &mut HashMap<String, u32>,
id2w: &mut Vec<String>,
p: &Option<ProgressBar>,
) -> (Vec<Word>, Vec<u32>) {
let mut words: Vec<Word> = Vec::with_capacity(wc.len());
let mut counts: Vec<u32> = Vec::with_capacity(wc.len());
for (word, count) in wc {
let mut current_word = Word::new();
counts.push(*count);
for (is_first, is_last, c) in word.chars().with_first_and_last() {
let mut s = c.to_string();
if w2id.contains_key(&s) {
// Found the initial char in the authorized alphabet
// Add the `continuing_subword_prefix` if relevant
if !is_first {
if let Some(prefix) = &self.continuing_subword_prefix {
s = format!("{}{}", prefix, s);
}
}
// Add the `end_of_word_suffix` if relevant
if is_last {
if let Some(suffix) = &self.end_of_word_suffix {
s = format!("{}{}", s, suffix);
}
}
// Insert the new formed string if necessary
if !w2id.contains_key(&s) {
id2w.push(s.clone());
w2id.insert(s.clone(), (id2w.len() - 1) as u32);
}
current_word.add(w2id[&s], 1); // We do not care about the len here
}
}
words.push(current_word);
if let Some(p) = p {
p.inc(1);
}
}
(words, counts)
}
fn count_pairs(
&self,
words: &[Word],
counts: &[u32],
p: &Option<ProgressBar>,
) -> (HashMap<Pair, i32>, HashMap<Pair, HashSet<usize>>) {
words
.maybe_par_iter()
.enumerate()
.map(|(i, word)| {
let mut pair_counts = HashMap::new();
let mut where_to_update: HashMap<Pair, HashSet<usize>> = HashMap::new();
for window in word.get_chars().windows(2) {
let cur_pair: Pair = (window[0], window[1]);
// Initialize pair_counts and where_to_update for this pair if we just saw it
if !pair_counts.contains_key(&cur_pair) {
pair_counts.insert(cur_pair, 0);
}
// Then update counts
let count = counts[i];
where_to_update
.entry(cur_pair)
.and_modify(|h| {
h.insert(i);
})
.or_insert_with(|| {
let mut h = HashSet::new();
h.insert(i);
h
});
*pair_counts.get_mut(&cur_pair).unwrap() += count as i32;
}
if let Some(p) = &p {
p.inc(1);
}
(pair_counts, where_to_update)
})
.reduce(
|| (HashMap::new(), HashMap::new()),
|(mut pair_counts, mut where_to_update), (pc, wtu)| {
for (k, v) in pc {
pair_counts.entry(k).and_modify(|c| *c += v).or_insert(v);
}
for (k, v) in wtu {
where_to_update
.entry(k)
.and_modify(|set| *set = set.union(&v).copied().collect())
.or_insert(v);
}
(pair_counts, where_to_update)
},
)
}
pub fn do_train(
&self,
word_counts: &HashMap<String, u32>,
model: &mut BPE,
) -> Result<Vec<AddedToken>> {
let mut word_to_id: HashMap<String, u32> = HashMap::with_capacity(self.vocab_size);
let mut id_to_word: Vec<String> = Vec::with_capacity(self.vocab_size);
let max_token_length: usize = self.max_token_length.unwrap_or(usize::MAX);
let progress = self.setup_progress();
//
// 1. Add all special tokens to the vocabulary
//
self.add_special_tokens(&mut word_to_id, &mut id_to_word);
//
// 2. Compute the initial alphabet
//
self.compute_alphabet(word_counts, &mut word_to_id, &mut id_to_word);
//
// 3. Tokenize words
//
self.update_progress(&progress, word_counts.len(), "Tokenize words");
let (words, counts) =
self.tokenize_words(word_counts, &mut word_to_id, &mut id_to_word, &progress);
self.finalize_progress(&progress, words.len());
//
// 4. Count pairs in words
//
self.update_progress(&progress, words.len(), "Count pairs");
let (mut pair_counts, mut where_to_update) = self.count_pairs(&words, &counts, &progress);
// Insert them in the queue
let mut queue = BinaryHeap::with_capacity(pair_counts.len());
where_to_update.drain().for_each(|(pair, pos)| {
let count = pair_counts[&pair];
if count > 0 {
queue.push(Merge {
pair,
count: count as u32,
pos,
});
}
});
self.finalize_progress(&progress, words.len());
//
// 5. Do merges
//
self.update_progress(&progress, self.vocab_size, "Compute merges");
let mut merges: Vec<(Pair, u32)> = vec![];
loop {
// Stop as soon as we have a big enough vocabulary
if word_to_id.len() >= self.vocab_size {
break;
}
if queue.is_empty() {
break;
}
let mut top = queue.pop().unwrap();
if top.count != pair_counts[&top.pair] as u32 {
top.count = pair_counts[&top.pair] as u32;
queue.push(top);
continue;
}
if top.count < 1 || self.min_frequency > top.count {
break;
}
let part_a = &id_to_word[top.pair.0 as usize];
let mut part_b = id_to_word[top.pair.1 as usize].to_owned();
// Build new token
if let Some(prefix) = &self.continuing_subword_prefix {
if part_b.starts_with(prefix) {
let prefix_byte_len = prefix.chars().map(|c| c.len_utf8()).sum();
part_b = part_b[prefix_byte_len..].to_string();
}
}
let new_token = format!("{}{}", part_a, part_b);
// implement sentencepiece-like merge.
// if this code were to be merged, integrate a way in the python bindings to communicate this variable
// default should be 0/None to maintain previous behavior. 16 is the spm default.
// Insert new token if it does not already exist
let new_token_id = word_to_id
.get(&new_token)
.copied()
.unwrap_or(id_to_word.len() as u32);
if word_to_id.get(&new_token).is_none() {
id_to_word.push(new_token.clone());
word_to_id.insert(new_token.clone(), new_token_id);
}
merges.push((top.pair, new_token_id));
// Merge the new pair in every words
let changes = top
.pos
.maybe_par_iter()
.flat_map(|&i| {
let word = &words[i] as *const _ as *mut Word;
// We can merge each of these words in parallel here because each position
// can be there only once (HashSet). So this is safe.
unsafe {
// let word: &mut Word = &mut (*word);
(*word)
.merge(top.pair.0, top.pair.1, new_token_id, max_token_length)
.into_iter()
.map(|c| (c, i))
.collect::<Vec<_>>()
}
})
.collect::<Vec<_>>();
// Introduce new formed pairs
for ((pair, change), iw) in changes {
let count = change * counts[iw] as i32;
pair_counts
.entry(pair)
.and_modify(|c| *c += count)
.or_insert(count);
if change > 0 {
where_to_update
.entry(pair)
.and_modify(|h| {
h.insert(iw);
})
.or_insert_with(|| {
let mut h = HashSet::new();
h.insert(iw);
h
});
}
}
where_to_update.drain().for_each(|(pair, pos)| {
let count = pair_counts[&pair];
if count > 0 {
queue.push(Merge {
pair,
count: count as u32,
pos,
});
}
});
if let Some(p) = &progress {
p.inc(1);
}
}
self.finalize_progress(&progress, merges.len());
// Transfer new vocab & options to model
model.vocab = word_to_id;
model.vocab_r = model
.vocab
.iter()
.map(|(key, val)| (*val, key.to_owned()))
.collect();
model.merges = merges
.into_iter()
.enumerate()
.map(|(i, (pair, new_token_id))| (pair, (i as u32, new_token_id)))
.collect();
if let Some(prefix) = &self.continuing_subword_prefix {
model.continuing_subword_prefix = Some(prefix.to_owned());
} else {
model.continuing_subword_prefix = None;
}
if let Some(suffix) = &self.end_of_word_suffix {
model.end_of_word_suffix = Some(suffix.to_owned());
} else {
model.end_of_word_suffix = None;
}
Ok(self.special_tokens.clone())
}
}
impl Trainer for BpeTrainer {
type Model = BPE;
/// Train a BPE model
fn train(&self, model: &mut BPE) -> Result<Vec<AddedToken>> {
self.do_train(&self.words, model)
}
/// Whether we should show progress
fn should_show_progress(&self) -> bool {
self.show_progress
}
fn feed<I, S, F>(&mut self, iterator: I, process: F) -> Result<()>
where
I: Iterator<Item = S> + Send,
S: AsRef<str> + Send,
F: Fn(&str) -> Result<Vec<String>> + Sync,
{
let words: Result<HashMap<String, u32>> = iterator
.maybe_par_bridge()
.map(|sequence| {
let words = process(sequence.as_ref())?;
let mut map = HashMap::new();
for word in words {
map.entry(word).and_modify(|c| *c += 1).or_insert(1);
}
Ok(map)
})
.reduce(
|| Ok(HashMap::new()),
|acc, ws| {
let mut acc = acc?;
for (k, v) in ws? {
acc.entry(k).and_modify(|c| *c += v).or_insert(v);
}
Ok(acc)
},
);
self.words = words?;
Ok(())
}
}
#[cfg(test)]
mod tests {
use super::{BpeTrainer, Pair, BPE};
use std::collections::HashMap;
#[test]
fn test_train() {
let word_counts: HashMap<String, u32> = [
("roses".into(), 1),
("are".into(), 2),
("red".into(), 1),
("voilets".into(), 1),
("blue".into(), 1),
("BERT".into(), 1),
("is".into(), 2),
("big".into(), 1),
("and".into(), 1),
("so".into(), 1),
("GPT-2".into(), 1),
]
.iter()
.cloned()
.collect();
let trainer = BpeTrainer::builder()
.show_progress(false)
.min_frequency(2)
.build();
let mut model = BPE::default();
trainer.do_train(&word_counts, &mut model).unwrap();
// Vocab should contain all of the characters from the `word_counts` mapping
// as well as three merges: 're', 'are', and 'is'.
let expected_vocab: HashMap<String, u32> = [
("-".into(), 0),
("2".into(), 1),
("B".into(), 2),
("E".into(), 3),
("G".into(), 4),
("P".into(), 5),
("R".into(), 6),
("T".into(), 7),
("a".into(), 8),
("b".into(), 9),
("d".into(), 10),
("e".into(), 11),
("g".into(), 12),
("i".into(), 13),
("l".into(), 14),
("n".into(), 15),
("o".into(), 16),
("r".into(), 17),
("s".into(), 18),
("t".into(), 19),
("u".into(), 20),
("v".into(), 21),
("re".into(), 22),
("are".into(), 23),
("is".into(), 24),
]
.iter()
.cloned()
.collect();
assert_eq!(model.vocab, expected_vocab);
// The keys in `merges` are pairs of symbols, the values are tuples of (rank, id),
// where 'rank' determines the order in which this merge will be applied during
// tokenization, and 'id' is the vocab id of the symbol resulting from merging
// the pair of symbols in the corresponding key.
let expected_merges: HashMap<Pair, (u32, u32)> = [
((17, 11), (0, 22)), // 'r' + 'e' -> 're'
((8, 22), (1, 23)), // 'a' + 're' -> 'are'
((13, 18), (2, 24)), // 'i' + 's' -> 'is'
]
.iter()
.cloned()
.collect();
assert_eq!(model.merges, expected_merges);
}
#[test]
fn bpe_test_max_token_length_16() {
/* bpe_test_max_token_length series of tests test the max_token_length flag of bpetrainer
// this is the more robust version that only tests max length of learned tokens
// (pre) tokenizer settings or vocab can be easily modified when necessary
*/
let max_token_length = 16;
let long_word_counts: HashMap<String, u32> = [
("singlelongtokenwithoutcasechange", 2),
("singleLongTokenWithCamelCaseChange", 2),
("Longsingletokenwithpunctu@t!onwithin", 2),
("Anotherlongsingletokenwithnumberw1th1n", 2),
("짧ìíêžë¬žììŽì§§ìí", 2), // korean 10 char
("ꞎíêžë¬žììŽêžŽíêžë¬žììŽêžŽíêžë¬ž", 2), // korean 16 char
("çå笊䞲çå笊䞲çå", 2), //simplified chinese 10 char
("é¿å笊䞲é¿å笊䞲é¿å笊䞲é¿å笊䞲", 2), // simp. chinese 16 char
("çãæååçãæåå", 2), // japanese 10 char
("é·ãæååé·ãæååé·ãæååé·", 2), // japanese 16 char
("so", 2),
("GPT-2", 2),
]
.iter()
.map(|(key, value)| (key.to_string(), *value))
.collect();
let trainer = BpeTrainer::builder()
.max_token_length(Some(max_token_length))
.show_progress(false)
.min_frequency(0)
.build();
let mut model = BPE::default();
trainer.do_train(&long_word_counts, &mut model).unwrap();
let vocab = model.get_vocab();
for token in vocab.keys() {
assert!(
token.chars().count() <= max_token_length,
"token too long : {} , chars().count() = {}",
token,
token.chars().count()
)
}
}
#[test]
fn bpe_test_max_token_length_direct_assert() {
/* more direct version of bpe_test_max_token_length test
// directly compares tokens with known expected values.
// maybe unstable depending on specific settings or changes.
*/
let long_word_counts: HashMap<String, u32> = [
("sin", 2),
("Sin", 2),
("Lon", 2),
("Ano", 2),
("짧ìí", 2),
("ꞎíêž", 2),
("çå笊", 2),
("é¿å笊", 2),
("çãæ", 2),
("é·ãæ", 2),
("so", 2),
("GP", 2),
]
.iter()
.map(|(key, value)| (key.to_string(), *value))
.collect();
let trainer = BpeTrainer::builder()
.max_token_length(Some(2))
.show_progress(false)
.min_frequency(0)
.build();
let mut model = BPE::default();
trainer.do_train(&long_word_counts, &mut model).unwrap();
let trained_vocab: HashMap<String, u32> = model.get_vocab();
let expected_vocab: HashMap<String, u32> = [
("ç", 12),
("n", 6),
("i", 5),
("s", 8),
("å笊", 23),
("é·", 14),
("ꞎ", 17),
("ãæ", 22),
("L", 2),
("in", 21),
("o", 7),
("ìí", 29),
("S", 4),
("P", 3),
("so", 27),
("笊", 13),
("æ", 11),
("å", 10),
("짧", 19),
("GP", 25),
("êž", 16),
("G", 1),
("An", 24),
("é¿", 15),
("A", 0),
("Lo", 26),
("ꞎí", 28),
("ã", 9),
("í", 20),
("ì", 18),
]
.iter()
.cloned()
.map(|(k, v)| (k.to_string(), v))
.collect();
assert_eq!(trained_vocab, expected_vocab)
}
}
| tokenizers/tokenizers/src/models/bpe/trainer.rs/0 | {
"file_path": "tokenizers/tokenizers/src/models/bpe/trainer.rs",
"repo_id": "tokenizers",
"token_count": 15117
} | 222 |
pub mod bert;
pub mod precompiled;
pub mod prepend;
pub mod replace;
pub mod strip;
pub mod unicode;
pub mod utils;
pub use crate::normalizers::bert::BertNormalizer;
pub use crate::normalizers::precompiled::Precompiled;
pub use crate::normalizers::prepend::Prepend;
pub use crate::normalizers::replace::Replace;
pub use crate::normalizers::strip::{Strip, StripAccents};
pub use crate::normalizers::unicode::{Nmt, NFC, NFD, NFKC, NFKD};
pub use crate::normalizers::utils::{Lowercase, Sequence};
use serde::{Deserialize, Serialize};
use crate::{NormalizedString, Normalizer};
/// Wrapper for known Normalizers.
#[derive(Clone, Debug, Deserialize, Serialize)]
#[serde(untagged)]
pub enum NormalizerWrapper {
BertNormalizer(BertNormalizer),
StripNormalizer(Strip),
StripAccents(StripAccents),
NFC(NFC),
NFD(NFD),
NFKC(NFKC),
NFKD(NFKD),
Sequence(Sequence),
Lowercase(Lowercase),
Nmt(Nmt),
Precompiled(Precompiled),
Replace(Replace),
Prepend(Prepend),
}
impl Normalizer for NormalizerWrapper {
fn normalize(&self, normalized: &mut NormalizedString) -> crate::Result<()> {
match self {
Self::BertNormalizer(bn) => bn.normalize(normalized),
Self::StripNormalizer(sn) => sn.normalize(normalized),
Self::StripAccents(sn) => sn.normalize(normalized),
Self::NFC(nfc) => nfc.normalize(normalized),
Self::NFD(nfd) => nfd.normalize(normalized),
Self::NFKC(nfkc) => nfkc.normalize(normalized),
Self::NFKD(nfkd) => nfkd.normalize(normalized),
Self::Sequence(sequence) => sequence.normalize(normalized),
Self::Lowercase(lc) => lc.normalize(normalized),
Self::Nmt(lc) => lc.normalize(normalized),
Self::Precompiled(lc) => lc.normalize(normalized),
Self::Replace(lc) => lc.normalize(normalized),
Self::Prepend(lc) => lc.normalize(normalized),
}
}
}
impl_enum_from!(BertNormalizer, NormalizerWrapper, BertNormalizer);
impl_enum_from!(NFKD, NormalizerWrapper, NFKD);
impl_enum_from!(NFKC, NormalizerWrapper, NFKC);
impl_enum_from!(NFC, NormalizerWrapper, NFC);
impl_enum_from!(NFD, NormalizerWrapper, NFD);
impl_enum_from!(Strip, NormalizerWrapper, StripNormalizer);
impl_enum_from!(StripAccents, NormalizerWrapper, StripAccents);
impl_enum_from!(Sequence, NormalizerWrapper, Sequence);
impl_enum_from!(Lowercase, NormalizerWrapper, Lowercase);
impl_enum_from!(Nmt, NormalizerWrapper, Nmt);
impl_enum_from!(Precompiled, NormalizerWrapper, Precompiled);
impl_enum_from!(Replace, NormalizerWrapper, Replace);
impl_enum_from!(Prepend, NormalizerWrapper, Prepend);
| tokenizers/tokenizers/src/normalizers/mod.rs/0 | {
"file_path": "tokenizers/tokenizers/src/normalizers/mod.rs",
"repo_id": "tokenizers",
"token_count": 1090
} | 223 |
mod pre_tokenizer;
mod scripts;
// Re-export the PreTokenizer
pub use pre_tokenizer::UnicodeScripts;
| tokenizers/tokenizers/src/pre_tokenizers/unicode_scripts/mod.rs/0 | {
"file_path": "tokenizers/tokenizers/src/pre_tokenizers/unicode_scripts/mod.rs",
"repo_id": "tokenizers",
"token_count": 35
} | 224 |
use std::borrow::Borrow;
use std::collections::HashMap;
use std::hash::Hash;
use std::sync::RwLock;
/// The default capacity for a `BPE`'s internal cache.
pub static DEFAULT_CACHE_CAPACITY: usize = 10_000;
/// Provides a simple multithread cache to speed up BPE tokenization that will try to read values
/// concurrently but won't block if another thread is writing.
/// The goal is clearly not the accuracy of the content, both get and set
/// are not guaranteed to actually get or set.
#[derive(Debug)]
pub(crate) struct Cache<K, V>
where
K: Eq + Hash + Clone,
V: Clone,
{
map: RwLock<HashMap<K, V>>,
pub capacity: usize,
}
// We dont really care about Cache comparison, so let's make them always equal
impl<K, V> PartialEq for Cache<K, V>
where
K: Eq + Hash + Clone,
V: Clone,
{
fn eq(&self, _other: &Cache<K, V>) -> bool {
true
}
}
impl<K, V> Default for Cache<K, V>
where
K: Eq + Hash + Clone,
V: Clone,
{
fn default() -> Self {
Self::new(DEFAULT_CACHE_CAPACITY)
}
}
impl<K, V> Cache<K, V>
where
K: Eq + Hash + Clone,
V: Clone,
{
/// Create new `Cache` with the given capacity.
pub(crate) fn new(capacity: usize) -> Self {
let map = RwLock::new(HashMap::with_capacity(capacity));
Cache { map, capacity }
}
/// Create a fresh `Cache` with the same configuration.
pub(crate) fn fresh(&self) -> Self {
Self::new(self.capacity)
}
/// Clear the cache.
pub(crate) fn clear(&self) {
self.map.write().unwrap().clear();
}
#[allow(dead_code)]
pub(crate) fn get_values<'a, I, Q>(&self, keys_iter: I) -> Option<Vec<Option<V>>>
where
I: Iterator<Item = &'a Q>,
K: Borrow<Q>,
Q: Hash + Eq + ?Sized + 'a,
{
if let Ok(ref mut cache) = self.map.try_read() {
Some(keys_iter.map(|k| cache.get(k).cloned()).collect())
} else {
None
}
}
pub(crate) fn get<Q>(&self, key: &Q) -> Option<V>
where
K: Borrow<Q>,
Q: Hash + Eq + ?Sized,
{
if let Ok(ref mut cache) = self.map.try_read() {
cache.get(key).cloned()
} else {
None
}
}
pub(crate) fn set_values<I>(&self, entries: I)
where
I: IntoIterator<Item = (K, V)>,
{
// Before trying to acquire a write lock, we check if we are already at
// capacity with a read handler.
if let Ok(cache) = self.map.try_read() {
if cache.len() >= self.capacity {
// At capacity, so do nothing.
return;
}
} else {
// If we couldn't acquire a read handle then we probably won't be able to acquire
// a write handle one quadrillionth of a second later.
return;
}
// Not at capacity, so try acquiring a write handle.
if let Ok(mut cache) = self.map.try_write() {
let free = self.capacity - cache.len();
cache.extend(entries.into_iter().take(free));
}
}
pub(crate) fn set(&self, key: K, value: V) {
self.set_values(std::iter::once((key, value)))
}
}
| tokenizers/tokenizers/src/utils/cache.rs/0 | {
"file_path": "tokenizers/tokenizers/src/utils/cache.rs",
"repo_id": "tokenizers",
"token_count": 1436
} | 225 |
use tokenizers::models::bpe::BPE;
use tokenizers::pre_tokenizers::whitespace::Whitespace;
use tokenizers::{DecoderWrapper, NormalizerWrapper, PostProcessorWrapper, PreTokenizerWrapper};
use tokenizers::{Model, Tokenizer, TokenizerBuilder};
#[test]
fn bpe_values_after_training() {
let mut tokenizer = TokenizerBuilder::<
BPE,
NormalizerWrapper,
PreTokenizerWrapper,
PostProcessorWrapper,
DecoderWrapper,
>::default()
.with_model(
BPE::builder()
.unk_token("[UNK]".to_string())
.dropout(0.1)
.build()
.unwrap(),
)
.build()
.unwrap();
let mut trainer = tokenizer.get_model().get_trainer();
tokenizer
.train_from_files(&mut trainer, vec!["./data/small.txt".to_string()])
.unwrap();
assert_eq!(tokenizer.get_model().dropout, Some(0.1));
assert_eq!(tokenizer.get_model().unk_token, Some("[UNK]".to_string()));
}
#[test]
fn bpe_continuing_subword_prefix_error() {
let mut tokenizer = TokenizerBuilder::<
BPE,
NormalizerWrapper,
PreTokenizerWrapper,
PostProcessorWrapper,
DecoderWrapper,
>::default()
.with_model(
BPE::builder()
.unk_token("[UNK]".to_string())
.continuing_subword_prefix("##".to_string())
.build()
.unwrap(),
)
.with_pre_tokenizer(Some(PreTokenizerWrapper::Whitespace(Whitespace {})))
.build()
.unwrap();
let mut trainer = tokenizer.get_model().get_trainer();
tokenizer
.train_from_files(&mut trainer, vec!["./data/small.txt".to_string()])
.unwrap();
tokenizer.save("tokenizer.json", true).unwrap();
let tokenizer = Tokenizer::from_file("tokenizer.json").unwrap();
assert_eq!(tokenizer.get_vocab_size(false), 1526);
std::fs::remove_file("tokenizer.json").unwrap();
}
| tokenizers/tokenizers/tests/training.rs/0 | {
"file_path": "tokenizers/tokenizers/tests/training.rs",
"repo_id": "tokenizers",
"token_count": 851
} | 226 |
# https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-23-11.html#rel-23-11
FROM nvcr.io/nvidia/pytorch:23.11-py3
LABEL maintainer="Hugging Face"
ARG DEBIAN_FRONTEND=noninteractive
ARG PYTORCH='2.1.0'
# Example: `cu102`, `cu113`, etc.
ARG CUDA='cu121'
RUN apt -y update
RUN apt install -y libaio-dev
RUN python3 -m pip install --no-cache-dir --upgrade pip
ARG REF=main
RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
RUN python3 -m pip uninstall -y torch torchvision torchaudio
# Install latest release PyTorch
# (PyTorch must be installed before pre-compiling any DeepSpeed c++/cuda ops.)
# (https://www.deepspeed.ai/tutorials/advanced-install/#pre-install-deepspeed-ops)
RUN python3 -m pip install --no-cache-dir -U torch==$PYTORCH torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/$CUDA
RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
# Uninstall `transformer-engine` shipped with the base image
RUN python3 -m pip uninstall -y transformer-engine
# Uninstall `torch-tensorrt` shipped with the base image
RUN python3 -m pip uninstall -y torch-tensorrt
# recompile apex
RUN python3 -m pip uninstall -y apex
# RUN git clone https://github.com/NVIDIA/apex
# `MAX_JOBS=1` disables parallel building to avoid cpu memory OOM when building image on GitHub Action (standard) runners
# TODO: check if there is alternative way to install latest apex
# RUN cd apex && MAX_JOBS=1 python3 -m pip install --global-option="--cpp_ext" --global-option="--cuda_ext" --no-cache -v --disable-pip-version-check .
# Pre-build **latest** DeepSpeed, so it would be ready for testing (otherwise, the 1st deepspeed test will timeout)
RUN python3 -m pip uninstall -y deepspeed
# This has to be run (again) inside the GPU VMs running the tests.
# The installation works here, but some tests fail, if we don't pre-build deepspeed again in the VMs running the tests.
# TODO: Find out why test fail.
RUN DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check 2>&1
# When installing in editable mode, `transformers` is not recognized as a package.
# this line must be added in order for python to be aware of transformers.
RUN cd transformers && python3 setup.py develop
# The base image ships with `pydantic==1.8.2` which is not working - i.e. the next command fails
RUN python3 -m pip install -U --no-cache-dir "pydantic<2"
RUN python3 -c "from deepspeed.launcher.runner import main"
| transformers/docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile/0 | {
"file_path": "transformers/docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile",
"repo_id": "transformers",
"token_count": 893
} | 227 |
<!--Copyright 2022 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# Wie konvertiert man ein ð€ Transformers-Modell in TensorFlow?
Die Tatsache, dass mehrere Frameworks fÃŒr die Verwendung mit ð€ Transformers zur VerfÃŒgung stehen, gibt Ihnen die FlexibilitÀt, deren StÀrken beim Entwurf Ihrer Anwendung auszuspielen.
Ihre Anwendung zu entwerfen, aber das bedeutet auch, dass die KompatibilitÀt fÌr jedes Modell einzeln hinzugefÌgt werden muss. Die gute Nachricht ist, dass
das HinzufÌgen von TensorFlow-KompatibilitÀt zu einem bestehenden Modell einfacher ist als [das HinzufÌgen eines neuen Modells von Grund auf](add_new_model)!
Ob Sie ein tieferes VerstÀndnis fÃŒr groÃe TensorFlow-Modelle haben möchten, einen wichtigen Open-Source-Beitrag leisten oder
TensorFlow fÃŒr das Modell Ihrer Wahl aktivieren wollen, dieser Leitfaden ist fÃŒr Sie.
Dieser Leitfaden befÀhigt Sie, ein Mitglied unserer Gemeinschaft, TensorFlow-Modellgewichte und/oder
Architekturen beizusteuern, die in ð€ Transformers verwendet werden sollen, und zwar mit minimaler Betreuung durch das Hugging Face Team. Das Schreiben eines neuen Modells
ist keine Kleinigkeit, aber ich hoffe, dass dieser Leitfaden dazu beitrÀgt, dass es weniger eine Achterbahnfahrt ð¢ und mehr ein Spaziergang im Park ð¶ ist.
Die Nutzung unserer kollektiven Erfahrungen ist absolut entscheidend, um diesen Prozess immer einfacher zu machen, und deshalb möchten wir
ermutigen Sie daher, VerbesserungsvorschlÀge fÌr diesen Leitfaden zu machen!
Bevor Sie tiefer eintauchen, empfehlen wir Ihnen, die folgenden Ressourcen zu lesen, wenn Sie neu in ð€ Transformers sind:
- [Allgemeiner Ãberblick ÃŒber ð€ Transformers](add_new_model#general-overview-of-transformers)
- [Die TensorFlow-Philosophie von Hugging Face](https://huggingface.co/blog/tensorflow-philosophy)
Im Rest dieses Leitfadens werden Sie lernen, was nötig ist, um eine neue TensorFlow Modellarchitektur hinzuzufÌgen, die
Verfahren zur Konvertierung von PyTorch in TensorFlow-Modellgewichte und wie Sie Unstimmigkeiten zwischen ML
Frameworks. Legen Sie los!
<Tip>
Sind Sie unsicher, ob das Modell, das Sie verwenden möchten, bereits eine entsprechende TensorFlow-Architektur hat?
ÃberprÃŒfen Sie das Feld `model_type` in der `config.json` des Modells Ihrer Wahl
([Beispiel](https://huggingface.co/bert-base-uncased/blob/main/config.json#L14)). Wenn der entsprechende Modellordner in
ð€ Transformers eine Datei hat, deren Name mit "modeling_tf" beginnt, bedeutet dies, dass es eine entsprechende TensorFlow
Architektur hat ([Beispiel](https://github.com/huggingface/transformers/tree/main/src/transformers/models/bert)).
</Tip>
## Schritt-fÃŒr-Schritt-Anleitung zum HinzufÃŒgen von TensorFlow-Modellarchitektur-Code
Es gibt viele Möglichkeiten, eine groÃe Modellarchitektur zu entwerfen, und viele Möglichkeiten, diesen Entwurf zu implementieren. Wie auch immer,
Sie erinnern sich vielleicht an unseren [allgemeinen Ãberblick ÃŒber ð€ Transformers](add_new_model#general-overview-of-transformers)
wissen, dass wir ein meinungsfreudiger Haufen sind - die Benutzerfreundlichkeit von ð€ Transformers hÀngt von konsistenten Designentscheidungen ab. Aus
Erfahrung können wir Ihnen ein paar wichtige Dinge Ìber das HinzufÌgen von TensorFlow-Modellen sagen:
- Erfinden Sie das Rad nicht neu! In den meisten FÀllen gibt es mindestens zwei Referenzimplementierungen, die Sie ÌberprÌfen sollten: das
PyTorch-Ãquivalent des Modells, das Sie implementieren, und andere TensorFlow-Modelle fÃŒr dieselbe Klasse von Problemen.
- Gute Modellimplementierungen ÃŒberleben den Test der Zeit. Dies geschieht nicht, weil der Code hÃŒbsch ist, sondern eher
sondern weil der Code klar, einfach zu debuggen und darauf aufzubauen ist. Wenn Sie den Maintainern das Leben mit Ihrer
TensorFlow-Implementierung leicht machen, indem Sie die gleichen Muster wie in anderen TensorFlow-Modellen nachbilden und die Abweichung
zur PyTorch-Implementierung minimieren, stellen Sie sicher, dass Ihr Beitrag lange Bestand haben wird.
- Bitten Sie um Hilfe, wenn Sie nicht weiterkommen! Das ð€ Transformers-Team ist da, um zu helfen, und wir haben wahrscheinlich Lösungen fÃŒr die gleichen
Probleme gefunden, vor denen Sie stehen.
Hier finden Sie einen Ãberblick ÃŒber die Schritte, die zum HinzufÃŒgen einer TensorFlow-Modellarchitektur erforderlich sind:
1. WÀhlen Sie das Modell, das Sie konvertieren möchten
2. Bereiten Sie die Transformers-Entwicklungsumgebung vor.
3. (Optional) Verstehen Sie die theoretischen Aspekte und die bestehende Implementierung
4. Implementieren Sie die Modellarchitektur
5. Implementieren Sie Modelltests
6. Reichen Sie den Pull-Antrag ein
7. (Optional) Erstellen Sie Demos und teilen Sie diese mit der Welt
### 1.-3. Bereiten Sie Ihren Modellbeitrag vor
**1. WÀhlen Sie das Modell, das Sie konvertieren möchten**
Beginnen wir mit den Grundlagen: Als erstes mÌssen Sie die Architektur kennen, die Sie konvertieren möchten. Wenn Sie
Sie sich nicht auf eine bestimmte Architektur festgelegt haben, ist es eine gute Möglichkeit, das ð€ Transformers-Team um VorschlÀge zu bitten.
Wir werden Sie zu den wichtigsten Architekturen fÃŒhren, die auf der TensorFlow-Seite noch fehlen.
Seite fehlen. Wenn das spezifische Modell, das Sie mit TensorFlow verwenden möchten, bereits eine Implementierung der TensorFlow-Architektur in
ð€ Transformers, aber es fehlen Gewichte, können Sie direkt in den
Abschnitt [Gewichtskonvertierung](#adding-tensorflow-weights-to-hub)
auf dieser Seite.
Der Einfachheit halber wird im Rest dieser Anleitung davon ausgegangen, dass Sie sich entschieden haben, mit der TensorFlow-Version von
*BrandNewBert* (dasselbe Beispiel wie in der [Anleitung](add_new_model), um ein neues Modell von Grund auf hinzuzufÃŒgen).
<Tip>
Bevor Sie mit der Arbeit an einer TensorFlow-Modellarchitektur beginnen, sollten Sie sich vergewissern, dass es keine laufenden BemÃŒhungen in dieser Richtung gibt.
Sie können nach `BrandNewBert` auf der
[pull request GitHub page](https://github.com/huggingface/transformers/pulls?q=is%3Apr), um zu bestÀtigen, dass es keine
TensorFlow-bezogene Pull-Anfrage gibt.
</Tip>
**2. Transformers-Entwicklungsumgebung vorbereiten**
Nachdem Sie die Modellarchitektur ausgewÀhlt haben, öffnen Sie einen PR-Entwurf, um Ihre Absicht zu signalisieren, daran zu arbeiten. Folgen Sie den
Anweisungen, um Ihre Umgebung einzurichten und einen PR-Entwurf zu öffnen.
1. Forken Sie das [repository](https://github.com/huggingface/transformers), indem Sie auf der Seite des Repositorys auf die SchaltflÀche 'Fork' klicken.
Seite des Repositorys klicken. Dadurch wird eine Kopie des Codes unter Ihrem GitHub-Benutzerkonto erstellt.
2. Klonen Sie Ihren `transformers` Fork auf Ihre lokale Festplatte und fÃŒgen Sie das Basis-Repository als Remote hinzu:
```bash
git clone https://github.com/[your Github handle]/transformers.git
cd transformers
git remote add upstream https://github.com/huggingface/transformers.git
```
3. Richten Sie eine Entwicklungsumgebung ein, indem Sie z.B. den folgenden Befehl ausfÃŒhren:
```bash
python -m venv .env
source .env/bin/activate
pip install -e ".[dev]"
```
AbhÀngig von Ihrem Betriebssystem und da die Anzahl der optionalen AbhÀngigkeiten von Transformers wÀchst, kann es sein, dass Sie bei diesem Befehl einen
Fehler mit diesem Befehl erhalten. Wenn das der Fall ist, stellen Sie sicher, dass Sie TensorFlow installieren und dann ausfÃŒhren:
```bash
pip install -e ".[quality]"
```
**Hinweis:** Sie mÃŒssen CUDA nicht installiert haben. Es reicht aus, das neue Modell auf der CPU laufen zu lassen.
4. Erstellen Sie eine Verzweigung mit einem beschreibenden Namen von Ihrer Hauptverzweigung
```bash
git checkout -b add_tf_brand_new_bert
```
5. Abrufen und zurÃŒcksetzen auf die aktuelle Hauptversion
```bash
git fetch upstream
git rebase upstream/main
```
6. FÃŒgen Sie eine leere `.py` Datei in `transformers/src/models/brandnewbert/` mit dem Namen `modeling_tf_brandnewbert.py` hinzu. Dies wird
Ihre TensorFlow-Modelldatei sein.
7. Ãbertragen Sie die Ãnderungen auf Ihr Konto mit:
```bash
git add .
git commit -m "initial commit"
git push -u origin add_tf_brand_new_bert
```
8. Wenn Sie zufrieden sind, gehen Sie auf die Webseite Ihrer Abspaltung auf GitHub. Klicken Sie auf "Pull request". Stellen Sie sicher, dass Sie das
GitHub-Handle einiger Mitglieder des Hugging Face-Teams als Reviewer hinzuzufÃŒgen, damit das Hugging Face-Team ÃŒber zukÃŒnftige Ãnderungen informiert wird.
zukÃŒnftige Ãnderungen benachrichtigt wird.
9. Ãndern Sie den PR in einen Entwurf, indem Sie auf der rechten Seite der GitHub-Pull-Request-Webseite auf "In Entwurf umwandeln" klicken.
Jetzt haben Sie eine Entwicklungsumgebung eingerichtet, um *BrandNewBert* nach TensorFlow in ð€ Transformers zu portieren.
**3. (Optional) Verstehen Sie die theoretischen Aspekte und die bestehende Implementierung**
Sie sollten sich etwas Zeit nehmen, um die Arbeit von *BrandNewBert* zu lesen, falls eine solche Beschreibung existiert. Möglicherweise gibt es groÃe
Abschnitte des Papiers, die schwer zu verstehen sind. Wenn das der Fall ist, ist das in Ordnung - machen Sie sich keine Sorgen! Das Ziel ist
ist es nicht, ein tiefes theoretisches VerstÀndnis des Papiers zu erlangen, sondern die notwendigen Informationen zu extrahieren, um
das Modell mit Hilfe von TensorFlow effektiv in ð€ Transformers neu zu implementieren. Das heiÃt, Sie mÃŒssen nicht zu viel Zeit auf die
viel Zeit auf die theoretischen Aspekte verwenden, sondern sich lieber auf die praktischen Aspekte konzentrieren, nÀmlich auf die bestehende Modelldokumentation
Seite (z.B. [model docs for BERT](model_doc/bert)).
Nachdem Sie die Grundlagen der Modelle, die Sie implementieren wollen, verstanden haben, ist es wichtig, die bestehende
Implementierung zu verstehen. Dies ist eine gute Gelegenheit, sich zu vergewissern, dass eine funktionierende Implementierung mit Ihren Erwartungen an das
Modell entspricht, und um technische Herausforderungen auf der TensorFlow-Seite vorauszusehen.
Es ist ganz natÌrlich, dass Sie sich von der Menge an Informationen, die Sie gerade aufgesogen haben, ÌberwÀltigt fÌhlen. Es ist
Es ist definitiv nicht erforderlich, dass Sie in dieser Phase alle Facetten des Modells verstehen. Dennoch empfehlen wir Ihnen dringend
ermutigen wir Sie, alle dringenden Fragen in unserem [Forum](https://discuss.huggingface.co/) zu klÀren.
### 4. Implementierung des Modells
Jetzt ist es an der Zeit, endlich mit dem Programmieren zu beginnen. Als Ausgangspunkt empfehlen wir die PyTorch-Datei selbst: Kopieren Sie den Inhalt von
modeling_brand_new_bert.py` in `src/transformers/models/brand_new_bert/` nach
modeling_tf_brand_new_bert.py`. Das Ziel dieses Abschnitts ist es, die Datei zu Àndern und die Importstruktur von
ð€ Transformers zu aktualisieren, so dass Sie `TFBrandNewBert` und
`TFBrandNewBert.from_pretrained(model_repo, from_pt=True)` erfolgreich ein funktionierendes TensorFlow *BrandNewBert* Modell lÀdt.
Leider gibt es kein Rezept, um ein PyTorch-Modell in TensorFlow zu konvertieren. Sie können jedoch unsere Auswahl an
Tipps befolgen, um den Prozess so reibungslos wie möglich zu gestalten:
- Stellen Sie `TF` dem Namen aller Klassen voran (z.B. wird `BrandNewBert` zu `TFBrandNewBert`).
- Die meisten PyTorch-Operationen haben einen direkten TensorFlow-Ersatz. Zum Beispiel entspricht `torch.nn.Linear` der Klasse
`tf.keras.layers.Dense`, `torch.nn.Dropout` entspricht `tf.keras.layers.Dropout`, usw. Wenn Sie sich nicht sicher sind
Ìber eine bestimmte Operation nicht sicher sind, können Sie die [TensorFlow-Dokumentation](https://www.tensorflow.org/api_docs/python/tf)
oder die [PyTorch-Dokumentation](https://pytorch.org/docs/stable/).
- Suchen Sie nach Mustern in der Codebasis von ð€ Transformers. Wenn Sie auf eine bestimmte Operation stoÃen, fÃŒr die es keinen direkten Ersatz gibt
Ersatz hat, stehen die Chancen gut, dass jemand anderes bereits das gleiche Problem hatte.
- Behalten Sie standardmÀÃig die gleichen Variablennamen und die gleiche Struktur wie in PyTorch bei. Dies erleichtert die Fehlersuche, die Verfolgung von
Probleme zu verfolgen und spÀtere Korrekturen vorzunehmen.
- Einige Ebenen haben in jedem Framework unterschiedliche Standardwerte. Ein bemerkenswertes Beispiel ist die Schicht fÃŒr die Batch-Normalisierung
epsilon (`1e-5` in [PyTorch](https://pytorch.org/docs/stable/generated/torch.nn.BatchNorm2d.html#torch.nn.BatchNorm2d)
und `1e-3` in [TensorFlow](https://www.tensorflow.org/api_docs/python/tf/keras/layers/BatchNormalization)).
PrÃŒfen Sie die Dokumentation genau!
- Die Variablen `nn.Parameter` von PyTorch mÃŒssen in der Regel innerhalb von TF Layer's `build()` initialisiert werden. Siehe das folgende
Beispiel: [PyTorch](https://github.com/huggingface/transformers/blob/655f72a6896c0533b1bdee519ed65a059c2425ac/src/transformers/models/vit_mae/modeling_vit_mae.py#L212) /
[TensorFlow](https://github.com/huggingface/transformers/blob/655f72a6896c0533b1bdee519ed65a059c2425ac/src/transformers/models/vit_mae/modeling_tf_vit_mae.py#L220)
- Wenn das PyTorch-Modell ein `#copied from ...` am Anfang einer Funktion hat, stehen die Chancen gut, dass Ihr TensorFlow-Modell diese Funktion auch
diese Funktion von der Architektur ausleihen kann, von der sie kopiert wurde, vorausgesetzt, es hat eine TensorFlow-Architektur.
- Die korrekte Zuweisung des Attributs `name` in TensorFlow-Funktionen ist entscheidend, um das `from_pt=True` Gewicht zu erreichen
Cross-Loading. Name" ist fast immer der Name der entsprechenden Variablen im PyTorch-Code. Wenn `name` nicht
nicht richtig gesetzt ist, sehen Sie dies in der Fehlermeldung beim Laden der Modellgewichte.
- Die Logik der Basismodellklasse, `BrandNewBertModel`, befindet sich in `TFBrandNewBertMainLayer`, einer Keras
Schicht-Unterklasse ([Beispiel](https://github.com/huggingface/transformers/blob/4fd32a1f499e45f009c2c0dea4d81c321cba7e02/src/transformers/models/bert/modeling_tf_bert.py#L719)).
TFBrandNewBertModel" ist lediglich ein Wrapper fÃŒr diese Schicht.
- Keras-Modelle mÃŒssen erstellt werden, um die vorher trainierten Gewichte zu laden. Aus diesem Grund muss `TFBrandNewBertPreTrainedModel`
ein Beispiel fÃŒr die Eingaben in das Modell enthalten, die `dummy_inputs`
([Beispiel](https://github.com/huggingface/transformers/blob/4fd32a1f499e45f009c2c0dea4d81c321cba7e02/src/transformers/models/bert/modeling_tf_bert.py#L916)).
- Wenn Sie nicht weiterkommen, fragen Sie nach Hilfe - wir sind fÃŒr Sie da! ð€
Neben der Modelldatei selbst mÌssen Sie auch die Verweise auf die Modellklassen und die zugehörigen
Dokumentationsseiten hinzufÌgen. Sie können diesen Teil ganz nach den Mustern in anderen PRs erledigen
([Beispiel](https://github.com/huggingface/transformers/pull/18020/files)). Hier ist eine Liste der erforderlichen manuellen
Ãnderungen:
- FÌgen Sie alle öffentlichen Klassen von *BrandNewBert* in `src/transformers/__init__.py` ein.
- FÃŒgen Sie *BrandNewBert* Klassen zu den entsprechenden Auto Klassen in `src/transformers/models/auto/modeling_tf_auto.py` hinzu.
- FÌgen Sie die *BrandNewBert* zugehörigen Klassen fÌr trÀges Laden in `src/transformers/utils/dummy_tf_objects.py` hinzu.
- Aktualisieren Sie die Importstrukturen fÌr die öffentlichen Klassen in `src/transformers/models/brand_new_bert/__init__.py`.
- FÌgen Sie die Dokumentationszeiger auf die öffentlichen Methoden von *BrandNewBert* in `docs/source/de/model_doc/brand_new_bert.md` hinzu.
- FÃŒgen Sie sich selbst zur Liste der Mitwirkenden an *BrandNewBert* in `docs/source/de/model_doc/brand_new_bert.md` hinzu.
- FÃŒgen Sie schlieÃlich ein grÃŒnes HÀkchen â
in der TensorFlow-Spalte von *BrandNewBert* in `docs/source/de/index.md` hinzu.
Wenn Sie mit Ihrer Implementierung zufrieden sind, fÌhren Sie die folgende Checkliste aus, um zu bestÀtigen, dass Ihre Modellarchitektur
fertig ist:
1. Alle Schichten, die sich zur Trainingszeit anders verhalten (z.B. Dropout), werden mit einem `Training` Argument aufgerufen, das
von den Top-Level-Klassen weitergegeben wird
2. Sie haben `#copied from ...` verwendet, wann immer es möglich war.
3. Die Funktion `TFBrandNewBertMainLayer` und alle Klassen, die sie verwenden, haben ihre Funktion `call` mit `@unpack_inputs` dekoriert
4. TFBrandNewBertMainLayer` ist mit `@keras_serializable` dekoriert
5. Ein TensorFlow-Modell kann aus PyTorch-Gewichten mit `TFBrandNewBert.from_pretrained(model_repo, from_pt=True)` geladen werden.
6. Sie können das TensorFlow Modell mit dem erwarteten Eingabeformat aufrufen
### 5. Modell-Tests hinzufÃŒgen
Hurra, Sie haben ein TensorFlow-Modell implementiert! Jetzt ist es an der Zeit, Tests hinzuzufÌgen, um sicherzustellen, dass sich Ihr Modell wie erwartet verhÀlt.
erwartet. Wie im vorigen Abschnitt schlagen wir vor, dass Sie zunÀchst die Datei `test_modeling_brand_new_bert.py` in
`tests/models/brand_new_bert/` in die Datei `test_modeling_tf_brand_new_bert.py` zu kopieren und dann die notwendigen
TensorFlow-Ersetzungen vornehmen. FÃŒr den Moment sollten Sie in allen Aufrufen von `.from_pretrained()` das Flag `from_pt=True` verwenden, um die
die vorhandenen PyTorch-Gewichte zu laden.
Wenn Sie damit fertig sind, kommt der Moment der Wahrheit: FÃŒhren Sie die Tests durch! ð¬
```bash
NVIDIA_TF32_OVERRIDE=0 RUN_SLOW=1 RUN_PT_TF_CROSS_TESTS=1 \
py.test -vv tests/models/brand_new_bert/test_modeling_tf_brand_new_bert.py
```
Das wahrscheinlichste Ergebnis ist, dass Sie eine Reihe von Fehlern sehen werden. Machen Sie sich keine Sorgen, das ist zu erwarten! Das Debuggen von ML-Modellen ist
notorisch schwierig, und der SchlÃŒssel zum Erfolg ist Geduld (und `breakpoint()`). Nach unserer Erfahrung sind die schwierigsten
Probleme aus subtilen Unstimmigkeiten zwischen ML-Frameworks, zu denen wir am Ende dieses Leitfadens ein paar Hinweise geben.
In anderen FÀllen kann es sein, dass ein allgemeiner Test nicht direkt auf Ihr Modell anwendbar ist; in diesem Fall empfehlen wir eine Ãberschreibung
auf der Ebene der Modelltestklasse. Zögern Sie nicht, in Ihrem Entwurf einer Pull-Anfrage um Hilfe zu bitten, wenn
Sie nicht weiterkommen.
Wenn alle Tests erfolgreich waren, können Sie Ihr Modell in die ð€ Transformers-Bibliothek aufnehmen! ð
### 6.-7. Stellen Sie sicher, dass jeder Ihr Modell verwenden kann
**6. Reichen Sie den Pull Request ein**
Sobald Sie mit der Implementierung und den Tests fertig sind, ist es an der Zeit, eine Pull-Anfrage einzureichen. Bevor Sie Ihren Code einreichen,
fÃŒhren Sie unser Dienstprogramm zur Codeformatierung, `make fixup` ðª, aus. Damit werden automatisch alle Formatierungsfehler behoben, die dazu fÃŒhren wÃŒrden, dass
unsere automatischen PrÃŒfungen fehlschlagen wÃŒrden.
Nun ist es an der Zeit, Ihren Entwurf einer Pull-Anfrage in eine echte Pull-Anfrage umzuwandeln. Klicken Sie dazu auf die SchaltflÀche "Bereit fÌr
Review" und fÌgen Sie Joao (`@gante`) und Matt (`@Rocketknight1`) als Reviewer hinzu. Eine Modell-Pull-Anfrage benötigt
mindestens 3 Reviewer, aber sie werden sich darum kÌmmern, geeignete zusÀtzliche Reviewer fÌr Ihr Modell zu finden.
Nachdem alle Gutachter mit dem Stand Ihres PR zufrieden sind, entfernen Sie als letzten Aktionspunkt das Flag `from_pt=True` in
.from_pretrained()-Aufrufen zu entfernen. Da es keine TensorFlow-Gewichte gibt, mÃŒssen Sie sie hinzufÃŒgen! Lesen Sie den Abschnitt
unten, um zu erfahren, wie Sie dies tun können.
Wenn schlieÃlich die TensorFlow-Gewichte zusammengefÃŒhrt werden, Sie mindestens 3 Genehmigungen von PrÃŒfern haben und alle CI-Checks grÃŒn sind
grÃŒn sind, ÃŒberprÃŒfen Sie die Tests ein letztes Mal lokal
```bash
NVIDIA_TF32_OVERRIDE=0 RUN_SLOW=1 RUN_PT_TF_CROSS_TESTS=1 \
py.test -vv tests/models/brand_new_bert/test_modeling_tf_brand_new_bert.py
```
und wir werden Ihren PR zusammenfÃŒhren! Herzlichen GlÃŒckwunsch zu dem Meilenstein ð.
**7. (Optional) Erstellen Sie Demos und teilen Sie sie mit der Welt**
Eine der schwierigsten Aufgaben bei Open-Source ist die Entdeckung. Wie können die anderen Benutzer von der Existenz Ihres
fabelhaften TensorFlow-Beitrags erfahren? Mit der richtigen Kommunikation, natÃŒrlich! ð£
Es gibt vor allem zwei Möglichkeiten, Ihr Modell mit der Community zu teilen:
- Erstellen Sie Demos. Dazu gehören Gradio-Demos, Notebooks und andere unterhaltsame Möglichkeiten, Ihr Modell vorzufÌhren. Wir raten Ihnen
ermutigen Sie, ein Notizbuch zu unseren [community-driven demos](https://huggingface.co/docs/transformers/community) hinzuzufÃŒgen.
- Teilen Sie Geschichten in sozialen Medien wie Twitter und LinkedIn. Sie sollten stolz auf Ihre Arbeit sein und sie mit der
Ihre Leistung mit der Community teilen - Ihr Modell kann nun von Tausenden von Ingenieuren und Forschern auf der ganzen Welt genutzt werden
der Welt genutzt werden ð! Wir werden Ihre BeitrÀge gerne retweeten und Ihnen helfen, Ihre Arbeit mit der Community zu teilen.
## HinzufÃŒgen von TensorFlow-Gewichten zum ð€ Hub
Unter der Annahme, dass die TensorFlow-Modellarchitektur in ð€ Transformers verfÃŒgbar ist, ist die Umwandlung von PyTorch-Gewichten in
TensorFlow-Gewichte ist ein Kinderspiel!
Hier sehen Sie, wie es geht:
1. Stellen Sie sicher, dass Sie in Ihrem Terminal bei Ihrem Hugging Face Konto angemeldet sind. Sie können sich mit dem folgenden Befehl anmelden
`huggingface-cli login` (Ihre Zugangstoken finden Sie [hier](https://huggingface.co/settings/tokens))
2. FÃŒhren Sie `transformers-cli pt-to-tf --model-name foo/bar` aus, wobei `foo/bar` der Name des Modell-Repositorys ist
ist, das die PyTorch-Gewichte enthÀlt, die Sie konvertieren möchten.
3. Markieren Sie `@joaogante` und `@Rocketknight1` in dem ð€ Hub PR, den der obige Befehl gerade erstellt hat
Das war's! ð
## Fehlersuche in verschiedenen ML-Frameworks ð
Irgendwann, wenn Sie eine neue Architektur hinzufÃŒgen oder TensorFlow-Gewichte fÃŒr eine bestehende Architektur erstellen, werden Sie
stoÃen Sie vielleicht auf Fehler, die sich ÃŒber Unstimmigkeiten zwischen PyTorch und TensorFlow beschweren. Sie könnten sich sogar dazu entschlieÃen, den
Modellarchitektur-Code fÃŒr die beiden Frameworks zu öffnen, und stellen fest, dass sie identisch aussehen. Was ist denn da los? ð€
Lassen Sie uns zunÀchst darÃŒber sprechen, warum es wichtig ist, diese Diskrepanzen zu verstehen. Viele Community-Mitglieder werden ð€
Transformers-Modelle und vertrauen darauf, dass sich unsere Modelle wie erwartet verhalten. Wenn es eine groÃe Diskrepanz gibt
zwischen den beiden Frameworks auftritt, bedeutet dies, dass das Modell nicht der Referenzimplementierung fÃŒr mindestens eines der Frameworks folgt.
der Frameworks folgt. Dies kann zu stillen Fehlern fÌhren, bei denen das Modell zwar lÀuft, aber eine schlechte Leistung aufweist. Dies ist
wohl schlimmer als ein Modell, das Ìberhaupt nicht lÀuft! Aus diesem Grund streben wir an, dass die Abweichung zwischen den Frameworks kleiner als
1e-5" in allen Phasen des Modells.
Wie bei anderen numerischen Problemen auch, steckt der Teufel im Detail. Und wie bei jedem detailorientierten Handwerk ist die geheime
Zutat hier Geduld. Hier ist unser Vorschlag fÃŒr den Arbeitsablauf, wenn Sie auf diese Art von Problemen stoÃen:
1. Lokalisieren Sie die Quelle der Abweichungen. Das Modell, das Sie konvertieren, hat wahrscheinlich bis zu einem gewissen Punkt nahezu identische innere Variablen.
bestimmten Punkt. Platzieren Sie `Breakpoint()`-Anweisungen in den Architekturen der beiden Frameworks und vergleichen Sie die Werte der
numerischen Variablen von oben nach unten, bis Sie die Quelle der Probleme gefunden haben.
2. Nachdem Sie nun die Ursache des Problems gefunden haben, setzen Sie sich mit dem ð€ Transformers-Team in Verbindung. Es ist möglich
dass wir ein Àhnliches Problem schon einmal gesehen haben und umgehend eine Lösung anbieten können. Als Ausweichmöglichkeit können Sie beliebte Seiten
wie StackOverflow und GitHub-Probleme.
3. Wenn keine Lösung in Sicht ist, bedeutet das, dass Sie tiefer gehen mÌssen. Die gute Nachricht ist, dass Sie das Problem gefunden haben.
Problem ausfindig gemacht haben, so dass Sie sich auf die problematische Anweisung konzentrieren und den Rest des Modells ausblenden können! Die schlechte Nachricht ist
dass Sie sich in die Quellimplementierung der besagten Anweisung einarbeiten mÌssen. In manchen FÀllen finden Sie vielleicht ein
Problem mit einer Referenzimplementierung - verzichten Sie nicht darauf, ein Problem im Upstream-Repository zu öffnen.
In einigen FÀllen können wir nach RÃŒcksprache mit dem ð€ Transformers-Team zu dem Schluss kommen, dass die Behebung der Abweichung nicht machbar ist.
Wenn die Abweichung in den Ausgabeschichten des Modells sehr klein ist (aber möglicherweise groà in den versteckten ZustÀnden), können wir
könnten wir beschlieÃen, sie zu ignorieren und das Modell zu verteilen. Die oben erwÀhnte CLI `pt-to-tf` hat ein `--max-error`
Flag, um die Fehlermeldung bei der Gewichtskonvertierung zu ÃŒberschreiben.
| transformers/docs/source/de/add_tensorflow_model.md/0 | {
"file_path": "transformers/docs/source/de/add_tensorflow_model.md",
"repo_id": "transformers",
"token_count": 9906
} | 228 |
# Optimizing inference
perf_infer_gpu_many: perf_infer_gpu_one
| transformers/docs/source/en/_redirects.yml/0 | {
"file_path": "transformers/docs/source/en/_redirects.yml",
"repo_id": "transformers",
"token_count": 25
} | 229 |
<!--Copyright 2023 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# Custom Tools and Prompts
<Tip>
If you are not aware of what tools and agents are in the context of transformers, we recommend you read the
[Transformers Agents](transformers_agents) page first.
</Tip>
<Tip warning={true}>
Transformers Agents is an experimental API that is subject to change at any time. Results returned by the agents
can vary as the APIs or underlying models are prone to change.
</Tip>
Creating and using custom tools and prompts is paramount to empowering the agent and having it perform new tasks.
In this guide we'll take a look at:
- How to customize the prompt
- How to use custom tools
- How to create custom tools
## Customizing the prompt
As explained in [Transformers Agents](transformers_agents) agents can run in [`~Agent.run`] and [`~Agent.chat`] mode.
Both the `run` and `chat` modes underlie the same logic. The language model powering the agent is conditioned on a long
prompt and completes the prompt by generating the next tokens until the stop token is reached.
The only difference between the two modes is that during the `chat` mode the prompt is extended with
previous user inputs and model generations. This allows the agent to have access to past interactions,
seemingly giving the agent some kind of memory.
### Structure of the prompt
Let's take a closer look at how the prompt is structured to understand how it can be best customized.
The prompt is structured broadly into four parts.
- 1. Introduction: how the agent should behave, explanation of the concept of tools.
- 2. Description of all the tools. This is defined by a `<<all_tools>>` token that is dynamically replaced at runtime with the tools defined/chosen by the user.
- 3. A set of examples of tasks and their solution
- 4. Current example, and request for solution.
To better understand each part, let's look at a shortened version of how the `run` prompt can look like:
````text
I will ask you to perform a task, your job is to come up with a series of simple commands in Python that will perform the task.
[...]
You can print intermediate results if it makes sense to do so.
Tools:
- document_qa: This is a tool that answers a question about a document (pdf). It takes an input named `document` which should be the document containing the information, as well as a `question` that is the question about the document. It returns a text that contains the answer to the question.
- image_captioner: This is a tool that generates a description of an image. It takes an input named `image` which should be the image to the caption and returns a text that contains the description in English.
[...]
Task: "Answer the question in the variable `question` about the image stored in the variable `image`. The question is in French."
I will use the following tools: `translator` to translate the question into English and then `image_qa` to answer the question on the input image.
Answer:
```py
translated_question = translator(question=question, src_lang="French", tgt_lang="English")
print(f"The translated question is {translated_question}.")
answer = image_qa(image=image, question=translated_question)
print(f"The answer is {answer}")
```
Task: "Identify the oldest person in the `document` and create an image showcasing the result as a banner."
I will use the following tools: `document_qa` to find the oldest person in the document, then `image_generator` to generate an image according to the answer.
Answer:
```py
answer = document_qa(document, question="What is the oldest person?")
print(f"The answer is {answer}.")
image = image_generator("A banner showing " + answer)
```
[...]
Task: "Draw me a picture of rivers and lakes"
I will use the following
````
The introduction (the text before *"Tools:"*) explains precisely how the model shall behave and what it should do.
This part most likely does not need to be customized as the agent shall always behave the same way.
The second part (the bullet points below *"Tools"*) is dynamically added upon calling `run` or `chat`. There are
exactly as many bullet points as there are tools in `agent.toolbox` and each bullet point consists of the name
and description of the tool:
```text
- <tool.name>: <tool.description>
```
Let's verify this quickly by loading the document_qa tool and printing out the name and description.
```py
from transformers import load_tool
document_qa = load_tool("document-question-answering")
print(f"- {document_qa.name}: {document_qa.description}")
```
which gives:
```text
- document_qa: This is a tool that answers a question about a document (pdf). It takes an input named `document` which should be the document containing the information, as well as a `question` that is the question about the document. It returns a text that contains the answer to the question.
```
We can see that the tool name is short and precise. The description includes two parts, the first explaining
what the tool does and the second states what input arguments and return values are expected.
A good tool name and tool description are very important for the agent to correctly use it. Note that the only
information the agent has about the tool is its name and description, so one should make sure that both
are precisely written and match the style of the existing tools in the toolbox. In particular make sure the description
mentions all the arguments expected by name in code-style, along with the expected type and a description of what they
are.
<Tip>
Check the naming and description of the curated Transformers tools to better understand what name and
description a tool is expected to have. You can see all tools with the [`Agent.toolbox`] property.
</Tip>
The third part includes a set of curated examples that show the agent exactly what code it should produce
for what kind of user request. The large language models empowering the agent are extremely good at
recognizing patterns in a prompt and repeating the pattern with new data. Therefore, it is very important
that the examples are written in a way that maximizes the likelihood of the agent to generating correct,
executable code in practice.
Let's have a look at one example:
````text
Task: "Identify the oldest person in the `document` and create an image showcasing the result as a banner."
I will use the following tools: `document_qa` to find the oldest person in the document, then `image_generator` to generate an image according to the answer.
Answer:
```py
answer = document_qa(document, question="What is the oldest person?")
print(f"The answer is {answer}.")
image = image_generator("A banner showing " + answer)
```
````
The pattern the model is prompted to repeat has three parts: The task statement, the agent's explanation of
what it intends to do, and finally the generated code. Every example that is part of the prompt has this exact
pattern, thus making sure that the agent will reproduce exactly the same pattern when generating new tokens.
The prompt examples are curated by the Transformers team and rigorously evaluated on a set of
[problem statements](https://github.com/huggingface/transformers/blob/main/src/transformers/tools/evaluate_agent.py)
to ensure that the agent's prompt is as good as possible to solve real use cases of the agent.
The final part of the prompt corresponds to:
```text
Task: "Draw me a picture of rivers and lakes"
I will use the following
```
is a final and unfinished example that the agent is tasked to complete. The unfinished example
is dynamically created based on the actual user input. For the above example, the user ran:
```py
agent.run("Draw me a picture of rivers and lakes")
```
The user input - *a.k.a* the task: *"Draw me a picture of rivers and lakes"* is cast into the
prompt template: "Task: <task> \n\n I will use the following". This sentence makes up the final lines of the
prompt the agent is conditioned on, therefore strongly influencing the agent to finish the example
exactly in the same way it was previously done in the examples.
Without going into too much detail, the chat template has the same prompt structure with the
examples having a slightly different style, *e.g.*:
````text
[...]
=====
Human: Answer the question in the variable `question` about the image stored in the variable `image`.
Assistant: I will use the tool `image_qa` to answer the question on the input image.
```py
answer = image_qa(text=question, image=image)
print(f"The answer is {answer}")
```
Human: I tried this code, it worked but didn't give me a good result. The question is in French
Assistant: In this case, the question needs to be translated first. I will use the tool `translator` to do this.
```py
translated_question = translator(question=question, src_lang="French", tgt_lang="English")
print(f"The translated question is {translated_question}.")
answer = image_qa(text=translated_question, image=image)
print(f"The answer is {answer}")
```
=====
[...]
````
Contrary, to the examples of the `run` prompt, each `chat` prompt example has one or more exchanges between the
*Human* and the *Assistant*. Every exchange is structured similarly to the example of the `run` prompt.
The user's input is appended to behind *Human:* and the agent is prompted to first generate what needs to be done
before generating code. An exchange can be based on previous exchanges, therefore allowing the user to refer
to past exchanges as is done *e.g.* above by the user's input of "I tried **this** code" refers to the
previously generated code of the agent.
Upon running `.chat`, the user's input or *task* is cast into an unfinished example of the form:
```text
Human: <user-input>\n\nAssistant:
```
which the agent completes. Contrary to the `run` command, the `chat` command then appends the completed example
to the prompt, thus giving the agent more context for the next `chat` turn.
Great now that we know how the prompt is structured, let's see how we can customize it!
### Writing good user inputs
While large language models are getting better and better at understanding users' intentions, it helps
enormously to be as precise as possible to help the agent pick the correct task. What does it mean to be
as precise as possible?
The agent sees a list of tool names and their description in its prompt. The more tools are added the
more difficult it becomes for the agent to choose the correct tool and it's even more difficult to choose
the correct sequences of tools to run. Let's look at a common failure case, here we will only return
the code to analyze it.
```py
from transformers import HfAgent
agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder")
agent.run("Show me a tree", return_code=True)
```
gives:
```text
==Explanation from the agent==
I will use the following tool: `image_segmenter` to create a segmentation mask for the image.
==Code generated by the agent==
mask = image_segmenter(image, prompt="tree")
```
which is probably not what we wanted. Instead, it is more likely that we want an image of a tree to be generated.
To steer the agent more towards using a specific tool it can therefore be very helpful to use important keywords that
are present in the tool's name and description. Let's have a look.
```py
agent.toolbox["image_generator"].description
```
```text
'This is a tool that creates an image according to a prompt, which is a text description. It takes an input named `prompt` which contains the image description and outputs an image.
```
The name and description make use of the keywords "image", "prompt", "create" and "generate". Using these words will most likely work better here. Let's refine our prompt a bit.
```py
agent.run("Create an image of a tree", return_code=True)
```
gives:
```text
==Explanation from the agent==
I will use the following tool `image_generator` to generate an image of a tree.
==Code generated by the agent==
image = image_generator(prompt="tree")
```
Much better! That looks more like what we want. In short, when you notice that the agent struggles to
correctly map your task to the correct tools, try looking up the most pertinent keywords of the tool's name
and description and try refining your task request with it.
### Customizing the tool descriptions
As we've seen before the agent has access to each of the tools' names and descriptions. The base tools
should have very precise names and descriptions, however, you might find that it could help to change the
the description or name of a tool for your specific use case. This might become especially important
when you've added multiple tools that are very similar or if you want to use your agent only for a certain
domain, *e.g.* image generation and transformations.
A common problem is that the agent confuses image generation with image transformation/modification when
used a lot for image generation tasks, *e.g.*
```py
agent.run("Make an image of a house and a car", return_code=True)
```
returns
```text
==Explanation from the agent==
I will use the following tools `image_generator` to generate an image of a house and `image_transformer` to transform the image of a car into the image of a house.
==Code generated by the agent==
house_image = image_generator(prompt="A house")
car_image = image_generator(prompt="A car")
house_car_image = image_transformer(image=car_image, prompt="A house")
```
which is probably not exactly what we want here. It seems like the agent has a difficult time
to understand the difference between `image_generator` and `image_transformer` and often uses the two together.
We can help the agent here by changing the tool name and description of `image_transformer`. Let's instead call it `modifier`
to disassociate it a bit from "image" and "prompt":
```py
agent.toolbox["modifier"] = agent.toolbox.pop("image_transformer")
agent.toolbox["modifier"].description = agent.toolbox["modifier"].description.replace(
"transforms an image according to a prompt", "modifies an image"
)
```
Now "modify" is a strong cue to use the new image processor which should help with the above prompt. Let's run it again.
```py
agent.run("Make an image of a house and a car", return_code=True)
```
Now we're getting:
```text
==Explanation from the agent==
I will use the following tools: `image_generator` to generate an image of a house, then `image_generator` to generate an image of a car.
==Code generated by the agent==
house_image = image_generator(prompt="A house")
car_image = image_generator(prompt="A car")
```
which is definitely closer to what we had in mind! However, we want to have both the house and car in the same image. Steering the task more toward single image generation should help:
```py
agent.run("Create image: 'A house and car'", return_code=True)
```
```text
==Explanation from the agent==
I will use the following tool: `image_generator` to generate an image.
==Code generated by the agent==
image = image_generator(prompt="A house and car")
```
<Tip warning={true}>
Agents are still brittle for many use cases, especially when it comes to
slightly more complex use cases like generating an image of multiple objects.
Both the agent itself and the underlying prompt will be further improved in the coming
months making sure that agents become more robust to a variety of user inputs.
</Tip>
### Customizing the whole prompt
To give the user maximum flexibility, the whole prompt template as explained in [above](#structure-of-the-prompt)
can be overwritten by the user. In this case make sure that your custom prompt includes an introduction section,
a tool section, an example section, and an unfinished example section. If you want to overwrite the `run` prompt template,
you can do as follows:
```py
template = """ [...] """
agent = HfAgent(your_endpoint, run_prompt_template=template)
```
<Tip warning={true}>
Please make sure to have the `<<all_tools>>` string and the `<<prompt>>` defined somewhere in the `template` so that the agent can be aware
of the tools, it has available to it as well as correctly insert the user's prompt.
</Tip>
Similarly, one can overwrite the `chat` prompt template. Note that the `chat` mode always uses the following format for the exchanges:
```text
Human: <<task>>
Assistant:
```
Therefore it is important that the examples of the custom `chat` prompt template also make use of this format.
You can overwrite the `chat` template at instantiation as follows.
```
template = """ [...] """
agent = HfAgent(url_endpoint=your_endpoint, chat_prompt_template=template)
```
<Tip warning={true}>
Please make sure to have the `<<all_tools>>` string defined somewhere in the `template` so that the agent can be aware
of the tools, it has available to it.
</Tip>
In both cases, you can pass a repo ID instead of the prompt template if you would like to use a template hosted by someone in the community. The default prompts live in [this repo](https://huggingface.co/datasets/huggingface-tools/default-prompts) as an example.
To upload your custom prompt on a repo on the Hub and share it with the community just make sure:
- to use a dataset repository
- to put the prompt template for the `run` command in a file named `run_prompt_template.txt`
- to put the prompt template for the `chat` command in a file named `chat_prompt_template.txt`
## Using custom tools
In this section, we'll be leveraging two existing custom tools that are specific to image generation:
- We replace [huggingface-tools/image-transformation](https://huggingface.co/spaces/huggingface-tools/image-transformation),
with [diffusers/controlnet-canny-tool](https://huggingface.co/spaces/diffusers/controlnet-canny-tool)
to allow for more image modifications.
- We add a new tool for image upscaling to the default toolbox:
[diffusers/latent-upscaler-tool](https://huggingface.co/spaces/diffusers/latent-upscaler-tool) replace the existing image-transformation tool.
We'll start by loading the custom tools with the convenient [`load_tool`] function:
```py
from transformers import load_tool
controlnet_transformer = load_tool("diffusers/controlnet-canny-tool")
upscaler = load_tool("diffusers/latent-upscaler-tool")
```
Upon adding custom tools to an agent, the tools' descriptions and names are automatically
included in the agents' prompts. Thus, it is imperative that custom tools have
a well-written description and name in order for the agent to understand how to use them.
Let's take a look at the description and name of `controlnet_transformer`:
```py
print(f"Description: '{controlnet_transformer.description}'")
print(f"Name: '{controlnet_transformer.name}'")
```
gives
```text
Description: 'This is a tool that transforms an image with ControlNet according to a prompt.
It takes two inputs: `image`, which should be the image to transform, and `prompt`, which should be the prompt to use to change it. It returns the modified image.'
Name: 'image_transformer'
```
The name and description are accurate and fit the style of the [curated set of tools](./transformers_agents#a-curated-set-of-tools).
Next, let's instantiate an agent with `controlnet_transformer` and `upscaler`:
```py
tools = [controlnet_transformer, upscaler]
agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder", additional_tools=tools)
```
This command should give you the following info:
```text
image_transformer has been replaced by <transformers_modules.diffusers.controlnet-canny-tool.bd76182c7777eba9612fc03c0
8718a60c0aa6312.image_transformation.ControlNetTransformationTool object at 0x7f1d3bfa3a00> as provided in `additional_tools`
```
The set of curated tools already has an `image_transformer` tool which is hereby replaced with our custom tool.
<Tip>
Overwriting existing tools can be beneficial if we want to use a custom tool exactly for the same task as an existing tool
because the agent is well-versed in using the specific task. Beware that the custom tool should follow the exact same API
as the overwritten tool in this case, or you should adapt the prompt template to make sure all examples using that
tool are updated.
</Tip>
The upscaler tool was given the name `image_upscaler` which is not yet present in the default toolbox and is therefore simply added to the list of tools.
You can always have a look at the toolbox that is currently available to the agent via the `agent.toolbox` attribute:
```py
print("\n".join([f"- {a}" for a in agent.toolbox.keys()]))
```
```text
- document_qa
- image_captioner
- image_qa
- image_segmenter
- transcriber
- summarizer
- text_classifier
- text_qa
- text_reader
- translator
- image_transformer
- text_downloader
- image_generator
- video_generator
- image_upscaler
```
Note how `image_upscaler` is now part of the agents' toolbox.
Let's now try out the new tools! We will re-use the image we generated in [Transformers Agents Quickstart](./transformers_agents#single-execution-run).
```py
from diffusers.utils import load_image
image = load_image(
"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png"
)
```
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200>
Let's transform the image into a beautiful winter landscape:
```py
image = agent.run("Transform the image: 'A frozen lake and snowy forest'", image=image)
```
```text
==Explanation from the agent==
I will use the following tool: `image_transformer` to transform the image.
==Code generated by the agent==
image = image_transformer(image, prompt="A frozen lake and snowy forest")
```
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes_winter.png" width=200>
The new image processing tool is based on ControlNet which can make very strong modifications to the image.
By default the image processing tool returns an image of size 512x512 pixels. Let's see if we can upscale it.
```py
image = agent.run("Upscale the image", image)
```
```text
==Explanation from the agent==
I will use the following tool: `image_upscaler` to upscale the image.
==Code generated by the agent==
upscaled_image = image_upscaler(image)
```
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes_winter_upscale.png" width=400>
The agent automatically mapped our prompt "Upscale the image" to the just added upscaler tool purely based on the description and name of the upscaler tool
and was able to correctly run it.
Next, let's have a look at how you can create a new custom tool.
### Adding new tools
In this section, we show how to create a new tool that can be added to the agent.
#### Creating a new tool
We'll first start by creating a tool. We'll add the not-so-useful yet fun task of fetching the model on the Hugging Face
Hub with the most downloads for a given task.
We can do that with the following code:
```python
from huggingface_hub import list_models
task = "text-classification"
model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
print(model.id)
```
For the task `text-classification`, this returns `'facebook/bart-large-mnli'`, for `translation` it returns `'t5-base`.
How do we convert this to a tool that the agent can leverage? All tools depend on the superclass `Tool` that holds the
main attributes necessary. We'll create a class that inherits from it:
```python
from transformers import Tool
class HFModelDownloadsTool(Tool):
pass
```
This class has a few needs:
- An attribute `name`, which corresponds to the name of the tool itself. To be in tune with other tools which have a
performative name, we'll name it `model_download_counter`.
- An attribute `description`, which will be used to populate the prompt of the agent.
- `inputs` and `outputs` attributes. Defining this will help the python interpreter make educated choices about types,
and will allow for a gradio-demo to be spawned when we push our tool to the Hub. They're both a list of expected
values, which can be `text`, `image`, or `audio`.
- A `__call__` method which contains the inference code. This is the code we've played with above!
Here's what our class looks like now:
```python
from transformers import Tool
from huggingface_hub import list_models
class HFModelDownloadsTool(Tool):
name = "model_download_counter"
description = (
"This is a tool that returns the most downloaded model of a given task on the Hugging Face Hub. "
"It takes the name of the category (such as text-classification, depth-estimation, etc), and "
"returns the name of the checkpoint."
)
inputs = ["text"]
outputs = ["text"]
def __call__(self, task: str):
model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
return model.id
```
We now have our tool handy. Save it in a file and import it from your main script. Let's name this file
`model_downloads.py`, so the resulting import code looks like this:
```python
from model_downloads import HFModelDownloadsTool
tool = HFModelDownloadsTool()
```
In order to let others benefit from it and for simpler initialization, we recommend pushing it to the Hub under your
namespace. To do so, just call `push_to_hub` on the `tool` variable:
```python
tool.push_to_hub("hf-model-downloads")
```
You now have your code on the Hub! Let's take a look at the final step, which is to have the agent use it.
#### Having the agent use the tool
We now have our tool that lives on the Hub which can be instantiated as such (change the user name for your tool):
```python
from transformers import load_tool
tool = load_tool("lysandre/hf-model-downloads")
```
In order to use it in the agent, simply pass it in the `additional_tools` parameter of the agent initialization method:
```python
from transformers import HfAgent
agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder", additional_tools=[tool])
agent.run(
"Can you read out loud the name of the model that has the most downloads in the 'text-to-video' task on the Hugging Face Hub?"
)
```
which outputs the following:
```text
==Code generated by the agent==
model = model_download_counter(task="text-to-video")
print(f"The model with the most downloads is {model}.")
audio_model = text_reader(model)
==Result==
The model with the most downloads is damo-vilab/text-to-video-ms-1.7b.
```
and generates the following audio.
| **Audio** |
|------------------------------------------------------------------------------------------------------------------------------------------------------|
| <audio controls><source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/damo.wav" type="audio/wav"/> |
<Tip>
Depending on the LLM, some are quite brittle and require very exact prompts in order to work well. Having a well-defined
name and description of the tool is paramount to having it be leveraged by the agent.
</Tip>
### Replacing existing tools
Replacing existing tools can be done simply by assigning a new item to the agent's toolbox. Here's how one would do so:
```python
from transformers import HfAgent, load_tool
agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder")
agent.toolbox["image-transformation"] = load_tool("diffusers/controlnet-canny-tool")
```
<Tip>
Beware when replacing tools with others! This will also adjust the agent's prompt. This can be good if you have a better
prompt suited for the task, but it can also result in your tool being selected way more than others or for other
tools to be selected instead of the one you have defined.
</Tip>
## Leveraging gradio-tools
[gradio-tools](https://github.com/freddyaboulton/gradio-tools) is a powerful library that allows using Hugging
Face Spaces as tools. It supports many existing Spaces as well as custom Spaces to be designed with it.
We offer support for `gradio_tools` by using the `Tool.from_gradio` method. For example, we want to take
advantage of the `StableDiffusionPromptGeneratorTool` tool offered in the `gradio-tools` toolkit so as to
improve our prompts and generate better images.
We first import the tool from `gradio_tools` and instantiate it:
```python
from gradio_tools import StableDiffusionPromptGeneratorTool
gradio_tool = StableDiffusionPromptGeneratorTool()
```
We pass that instance to the `Tool.from_gradio` method:
```python
from transformers import Tool
tool = Tool.from_gradio(gradio_tool)
```
Now we can manage it exactly as we would a usual custom tool. We leverage it to improve our prompt
` a rabbit wearing a space suit`:
```python
from transformers import HfAgent
agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder", additional_tools=[tool])
agent.run("Generate an image of the `prompt` after improving it.", prompt="A rabbit wearing a space suit")
```
The model adequately leverages the tool:
```text
==Explanation from the agent==
I will use the following tools: `StableDiffusionPromptGenerator` to improve the prompt, then `image_generator` to generate an image according to the improved prompt.
==Code generated by the agent==
improved_prompt = StableDiffusionPromptGenerator(prompt)
print(f"The improved prompt is {improved_prompt}.")
image = image_generator(improved_prompt)
```
Before finally generating the image:
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit.png">
<Tip warning={true}>
gradio-tools requires *textual* inputs and outputs, even when working with different modalities. This implementation
works with image and audio objects. The two are currently incompatible, but will rapidly become compatible as we
work to improve the support.
</Tip>
## Future compatibility with Langchain
We love Langchain and think it has a very compelling suite of tools. In order to handle these tools,
Langchain requires *textual* inputs and outputs, even when working with different modalities.
This is often the serialized version (i.e., saved to disk) of the objects.
This difference means that multi-modality isn't handled between transformers-agents and langchain.
We aim for this limitation to be resolved in future versions, and welcome any help from avid langchain
users to help us achieve this compatibility.
We would love to have better support. If you would like to help, please
[open an issue](https://github.com/huggingface/transformers/issues/new) and share what you have in mind.
| transformers/docs/source/en/custom_tools.md/0 | {
"file_path": "transformers/docs/source/en/custom_tools.md",
"repo_id": "transformers",
"token_count": 8660
} | 230 |
<!--Copyright 2020 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# Models
The base classes [`PreTrainedModel`], [`TFPreTrainedModel`], and
[`FlaxPreTrainedModel`] implement the common methods for loading/saving a model either from a local
file or directory, or from a pretrained model configuration provided by the library (downloaded from HuggingFace's AWS
S3 repository).
[`PreTrainedModel`] and [`TFPreTrainedModel`] also implement a few methods which
are common among all the models to:
- resize the input token embeddings when new tokens are added to the vocabulary
- prune the attention heads of the model.
The other methods that are common to each model are defined in [`~modeling_utils.ModuleUtilsMixin`]
(for the PyTorch models) and [`~modeling_tf_utils.TFModuleUtilsMixin`] (for the TensorFlow models) or
for text generation, [`~generation.GenerationMixin`] (for the PyTorch models),
[`~generation.TFGenerationMixin`] (for the TensorFlow models) and
[`~generation.FlaxGenerationMixin`] (for the Flax/JAX models).
## PreTrainedModel
[[autodoc]] PreTrainedModel
- push_to_hub
- all
<a id='from_pretrained-torch-dtype'></a>
### Large model loading
In Transformers 4.20.0, the [`~PreTrainedModel.from_pretrained`] method has been reworked to accommodate large models using [Accelerate](https://huggingface.co/docs/accelerate/big_modeling). This requires Accelerate >= 0.9.0 and PyTorch >= 1.9.0. Instead of creating the full model, then loading the pretrained weights inside it (which takes twice the size of the model in RAM, one for the randomly initialized model, one for the weights), there is an option to create the model as an empty shell, then only materialize its parameters when the pretrained weights are loaded.
This option can be activated with `low_cpu_mem_usage=True`. The model is first created on the Meta device (with empty weights) and the state dict is then loaded inside it (shard by shard in the case of a sharded checkpoint). This way the maximum RAM used is the full size of the model only.
```py
from transformers import AutoModelForSeq2SeqLM
t0pp = AutoModelForSeq2SeqLM.from_pretrained("bigscience/T0pp", low_cpu_mem_usage=True)
```
Moreover, you can directly place the model on different devices if it doesn't fully fit in RAM (only works for inference for now). With `device_map="auto"`, Accelerate will determine where to put each layer to maximize the use of your fastest devices (GPUs) and offload the rest on the CPU, or even the hard drive if you don't have enough GPU RAM (or CPU RAM). Even if the model is split across several devices, it will run as you would normally expect.
When passing a `device_map`, `low_cpu_mem_usage` is automatically set to `True`, so you don't need to specify it:
```py
from transformers import AutoModelForSeq2SeqLM
t0pp = AutoModelForSeq2SeqLM.from_pretrained("bigscience/T0pp", device_map="auto")
```
You can inspect how the model was split across devices by looking at its `hf_device_map` attribute:
```py
t0pp.hf_device_map
```
```python out
{'shared': 0,
'decoder.embed_tokens': 0,
'encoder': 0,
'decoder.block.0': 0,
'decoder.block.1': 1,
'decoder.block.2': 1,
'decoder.block.3': 1,
'decoder.block.4': 1,
'decoder.block.5': 1,
'decoder.block.6': 1,
'decoder.block.7': 1,
'decoder.block.8': 1,
'decoder.block.9': 1,
'decoder.block.10': 1,
'decoder.block.11': 1,
'decoder.block.12': 1,
'decoder.block.13': 1,
'decoder.block.14': 1,
'decoder.block.15': 1,
'decoder.block.16': 1,
'decoder.block.17': 1,
'decoder.block.18': 1,
'decoder.block.19': 1,
'decoder.block.20': 1,
'decoder.block.21': 1,
'decoder.block.22': 'cpu',
'decoder.block.23': 'cpu',
'decoder.final_layer_norm': 'cpu',
'decoder.dropout': 'cpu',
'lm_head': 'cpu'}
```
You can also write your own device map following the same format (a dictionary layer name to device). It should map all parameters of the model to a given device, but you don't have to detail where all the submodules of one layer go if that layer is entirely on the same device. For instance, the following device map would work properly for T0pp (as long as you have the GPU memory):
```python
device_map = {"shared": 0, "encoder": 0, "decoder": 1, "lm_head": 1}
```
Another way to minimize the memory impact of your model is to instantiate it at a lower precision dtype (like `torch.float16`) or use direct quantization techniques as described below.
### Model Instantiation dtype
Under Pytorch a model normally gets instantiated with `torch.float32` format. This can be an issue if one tries to
load a model whose weights are in fp16, since it'd require twice as much memory. To overcome this limitation, you can
either explicitly pass the desired `dtype` using `torch_dtype` argument:
```python
model = T5ForConditionalGeneration.from_pretrained("t5", torch_dtype=torch.float16)
```
or, if you want the model to always load in the most optimal memory pattern, you can use the special value `"auto"`,
and then `dtype` will be automatically derived from the model's weights:
```python
model = T5ForConditionalGeneration.from_pretrained("t5", torch_dtype="auto")
```
Models instantiated from scratch can also be told which `dtype` to use with:
```python
config = T5Config.from_pretrained("t5")
model = AutoModel.from_config(config)
```
Due to Pytorch design, this functionality is only available for floating dtypes.
## ModuleUtilsMixin
[[autodoc]] modeling_utils.ModuleUtilsMixin
## TFPreTrainedModel
[[autodoc]] TFPreTrainedModel
- push_to_hub
- all
## TFModelUtilsMixin
[[autodoc]] modeling_tf_utils.TFModelUtilsMixin
## FlaxPreTrainedModel
[[autodoc]] FlaxPreTrainedModel
- push_to_hub
- all
## Pushing to the Hub
[[autodoc]] utils.PushToHubMixin
## Sharded checkpoints
[[autodoc]] modeling_utils.load_sharded_checkpoint
| transformers/docs/source/en/main_classes/model.md/0 | {
"file_path": "transformers/docs/source/en/main_classes/model.md",
"repo_id": "transformers",
"token_count": 2010
} | 231 |
<!--Copyright 2023 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
-->
# Bark
## Overview
Bark is a transformer-based text-to-speech model proposed by Suno AI in [suno-ai/bark](https://github.com/suno-ai/bark).
Bark is made of 4 main models:
- [`BarkSemanticModel`] (also referred to as the 'text' model): a causal auto-regressive transformer model that takes as input tokenized text, and predicts semantic text tokens that capture the meaning of the text.
- [`BarkCoarseModel`] (also referred to as the 'coarse acoustics' model): a causal autoregressive transformer, that takes as input the results of the [`BarkSemanticModel`] model. It aims at predicting the first two audio codebooks necessary for EnCodec.
- [`BarkFineModel`] (the 'fine acoustics' model), this time a non-causal autoencoder transformer, which iteratively predicts the last codebooks based on the sum of the previous codebooks embeddings.
- having predicted all the codebook channels from the [`EncodecModel`], Bark uses it to decode the output audio array.
It should be noted that each of the first three modules can support conditional speaker embeddings to condition the output sound according to specific predefined voice.
This model was contributed by [Yoach Lacombe (ylacombe)](https://huggingface.co/ylacombe) and [Sanchit Gandhi (sanchit-gandhi)](https://github.com/sanchit-gandhi).
The original code can be found [here](https://github.com/suno-ai/bark).
### Optimizing Bark
Bark can be optimized with just a few extra lines of code, which **significantly reduces its memory footprint** and **accelerates inference**.
#### Using half-precision
You can speed up inference and reduce memory footprint by 50% simply by loading the model in half-precision.
```python
from transformers import BarkModel
import torch
device = "cuda" if torch.cuda.is_available() else "cpu"
model = BarkModel.from_pretrained("suno/bark-small", torch_dtype=torch.float16).to(device)
```
#### Using CPU offload
As mentioned above, Bark is made up of 4 sub-models, which are called up sequentially during audio generation. In other words, while one sub-model is in use, the other sub-models are idle.
If you're using a CUDA device, a simple solution to benefit from an 80% reduction in memory footprint is to offload the submodels from GPU to CPU when they're idle. This operation is called *CPU offloading*. You can use it with one line of code as follows:
```python
model.enable_cpu_offload()
```
Note that ð€ Accelerate must be installed before using this feature. [Here's how to install it.](https://huggingface.co/docs/accelerate/basic_tutorials/install)
#### Using Better Transformer
Better Transformer is an ð€ Optimum feature that performs kernel fusion under the hood. You can gain 20% to 30% in speed with zero performance degradation. It only requires one line of code to export the model to ð€ Better Transformer:
```python
model = model.to_bettertransformer()
```
Note that ð€ Optimum must be installed before using this feature. [Here's how to install it.](https://huggingface.co/docs/optimum/installation)
#### Using Flash Attention 2
Flash Attention 2 is an even faster, optimized version of the previous optimization.
##### Installation
First, check whether your hardware is compatible with Flash Attention 2. The latest list of compatible hardware can be found in the [official documentation](https://github.com/Dao-AILab/flash-attention#installation-and-features). If your hardware is not compatible with Flash Attention 2, you can still benefit from attention kernel optimisations through Better Transformer support covered [above](https://huggingface.co/docs/transformers/main/en/model_doc/bark#using-better-transformer).
Next, [install](https://github.com/Dao-AILab/flash-attention#installation-and-features) the latest version of Flash Attention 2:
```bash
pip install -U flash-attn --no-build-isolation
```
##### Usage
To load a model using Flash Attention 2, we can pass the `attn_implementation="flash_attention_2"` flag to [`.from_pretrained`](https://huggingface.co/docs/transformers/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained). We'll also load the model in half-precision (e.g. `torch.float16`), since it results in almost no degradation to audio quality but significantly lower memory usage and faster inference:
```python
model = BarkModel.from_pretrained("suno/bark-small", torch_dtype=torch.float16, attn_implementation="flash_attention_2").to(device)
```
##### Performance comparison
The following diagram shows the latency for the native attention implementation (no optimisation) against Better Transformer and Flash Attention 2. In all cases, we generate 400 semantic tokens on a 40GB A100 GPU with PyTorch 2.1. Flash Attention 2 is also consistently faster than Better Transformer, and its performance improves even more as batch sizes increase:
<div style="text-align: center">
<img src="https://huggingface.co/datasets/ylacombe/benchmark-comparison/resolve/main/Bark%20Optimization%20Benchmark.png">
</div>
To put this into perspective, on an NVIDIA A100 and when generating 400 semantic tokens with a batch size of 16, you can get 17 times the [throughput](https://huggingface.co/blog/optimizing-bark#throughput) and still be 2 seconds faster than generating sentences one by one with the native model implementation. In other words, all the samples will be generated 17 times faster.
At batch size 8, on an NVIDIA A100, Flash Attention 2 is also 10% faster than Better Transformer, and at batch size 16, 25%.
#### Combining optimization techniques
You can combine optimization techniques, and use CPU offload, half-precision and Flash Attention 2 (or ð€ Better Transformer) all at once.
```python
from transformers import BarkModel
import torch
device = "cuda" if torch.cuda.is_available() else "cpu"
# load in fp16 and use Flash Attention 2
model = BarkModel.from_pretrained("suno/bark-small", torch_dtype=torch.float16, attn_implementation="flash_attention_2").to(device)
# enable CPU offload
model.enable_cpu_offload()
```
Find out more on inference optimization techniques [here](https://huggingface.co/docs/transformers/perf_infer_gpu_one).
### Usage tips
Suno offers a library of voice presets in a number of languages [here](https://suno-ai.notion.site/8b8e8749ed514b0cbf3f699013548683?v=bc67cff786b04b50b3ceb756fd05f68c).
These presets are also uploaded in the hub [here](https://huggingface.co/suno/bark-small/tree/main/speaker_embeddings) or [here](https://huggingface.co/suno/bark/tree/main/speaker_embeddings).
```python
>>> from transformers import AutoProcessor, BarkModel
>>> processor = AutoProcessor.from_pretrained("suno/bark")
>>> model = BarkModel.from_pretrained("suno/bark")
>>> voice_preset = "v2/en_speaker_6"
>>> inputs = processor("Hello, my dog is cute", voice_preset=voice_preset)
>>> audio_array = model.generate(**inputs)
>>> audio_array = audio_array.cpu().numpy().squeeze()
```
Bark can generate highly realistic, **multilingual** speech as well as other audio - including music, background noise and simple sound effects.
```python
>>> # Multilingual speech - simplified Chinese
>>> inputs = processor("æ人çïŒæäŒè¯Žäžæ")
>>> # Multilingual speech - French - let's use a voice_preset as well
>>> inputs = processor("Incroyable! Je peux générer du son.", voice_preset="fr_speaker_5")
>>> # Bark can also generate music. You can help it out by adding music notes around your lyrics.
>>> inputs = processor("⪠Hello, my dog is cute âª")
>>> audio_array = model.generate(**inputs)
>>> audio_array = audio_array.cpu().numpy().squeeze()
```
The model can also produce **nonverbal communications** like laughing, sighing and crying.
```python
>>> # Adding non-speech cues to the input text
>>> inputs = processor("Hello uh ... [clears throat], my dog is cute [laughter]")
>>> audio_array = model.generate(**inputs)
>>> audio_array = audio_array.cpu().numpy().squeeze()
```
To save the audio, simply take the sample rate from the model config and some scipy utility:
```python
>>> from scipy.io.wavfile import write as write_wav
>>> # save audio to disk, but first take the sample rate from the model config
>>> sample_rate = model.generation_config.sample_rate
>>> write_wav("bark_generation.wav", sample_rate, audio_array)
```
## BarkConfig
[[autodoc]] BarkConfig
- all
## BarkProcessor
[[autodoc]] BarkProcessor
- all
- __call__
## BarkModel
[[autodoc]] BarkModel
- generate
- enable_cpu_offload
## BarkSemanticModel
[[autodoc]] BarkSemanticModel
- forward
## BarkCoarseModel
[[autodoc]] BarkCoarseModel
- forward
## BarkFineModel
[[autodoc]] BarkFineModel
- forward
## BarkCausalModel
[[autodoc]] BarkCausalModel
- forward
## BarkCoarseConfig
[[autodoc]] BarkCoarseConfig
- all
## BarkFineConfig
[[autodoc]] BarkFineConfig
- all
## BarkSemanticConfig
[[autodoc]] BarkSemanticConfig
- all
| transformers/docs/source/en/model_doc/bark.md/0 | {
"file_path": "transformers/docs/source/en/model_doc/bark.md",
"repo_id": "transformers",
"token_count": 2760
} | 232 |
<!--Copyright 2023 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# BLIP
## Overview
The BLIP model was proposed in [BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation](https://arxiv.org/abs/2201.12086) by Junnan Li, Dongxu Li, Caiming Xiong, Steven Hoi.
BLIP is a model that is able to perform various multi-modal tasks including:
- Visual Question Answering
- Image-Text retrieval (Image-text matching)
- Image Captioning
The abstract from the paper is the following:
*Vision-Language Pre-training (VLP) has advanced the performance for many vision-language tasks.
However, most existing pre-trained models only excel in either understanding-based tasks or generation-based tasks. Furthermore, performance improvement has been largely achieved by scaling up the dataset with noisy image-text pairs collected from the web, which is a suboptimal source of supervision. In this paper, we propose BLIP, a new VLP framework which transfers flexibly to both vision-language understanding and generation tasks. BLIP effectively utilizes the noisy web data by bootstrapping the captions, where a captioner generates synthetic captions and a filter removes the noisy ones. We achieve state-of-the-art results on a wide range of vision-language tasks, such as image-text retrieval (+2.7% in average recall@1), image captioning (+2.8% in CIDEr), and VQA (+1.6% in VQA score). BLIP also demonstrates strong generalization ability when directly transferred to videolanguage tasks in a zero-shot manner. Code, models, and datasets are released.*
This model was contributed by [ybelkada](https://huggingface.co/ybelkada).
The original code can be found [here](https://github.com/salesforce/BLIP).
## Resources
- [Jupyter notebook](https://github.com/huggingface/notebooks/blob/main/examples/image_captioning_blip.ipynb) on how to fine-tune BLIP for image captioning on a custom dataset
## BlipConfig
[[autodoc]] BlipConfig
- from_text_vision_configs
## BlipTextConfig
[[autodoc]] BlipTextConfig
## BlipVisionConfig
[[autodoc]] BlipVisionConfig
## BlipProcessor
[[autodoc]] BlipProcessor
## BlipImageProcessor
[[autodoc]] BlipImageProcessor
- preprocess
<frameworkcontent>
<pt>
## BlipModel
[[autodoc]] BlipModel
- forward
- get_text_features
- get_image_features
## BlipTextModel
[[autodoc]] BlipTextModel
- forward
## BlipVisionModel
[[autodoc]] BlipVisionModel
- forward
## BlipForConditionalGeneration
[[autodoc]] BlipForConditionalGeneration
- forward
## BlipForImageTextRetrieval
[[autodoc]] BlipForImageTextRetrieval
- forward
## BlipForQuestionAnswering
[[autodoc]] BlipForQuestionAnswering
- forward
</pt>
<tf>
## TFBlipModel
[[autodoc]] TFBlipModel
- call
- get_text_features
- get_image_features
## TFBlipTextModel
[[autodoc]] TFBlipTextModel
- call
## TFBlipVisionModel
[[autodoc]] TFBlipVisionModel
- call
## TFBlipForConditionalGeneration
[[autodoc]] TFBlipForConditionalGeneration
- call
## TFBlipForImageTextRetrieval
[[autodoc]] TFBlipForImageTextRetrieval
- call
## TFBlipForQuestionAnswering
[[autodoc]] TFBlipForQuestionAnswering
- call
</tf>
</frameworkcontent>
| transformers/docs/source/en/model_doc/blip.md/0 | {
"file_path": "transformers/docs/source/en/model_doc/blip.md",
"repo_id": "transformers",
"token_count": 1242
} | 233 |
<!--Copyright 2021 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# DETR
## Overview
The DETR model was proposed in [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by
Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov and Sergey Zagoruyko. DETR
consists of a convolutional backbone followed by an encoder-decoder Transformer which can be trained end-to-end for
object detection. It greatly simplifies a lot of the complexity of models like Faster-R-CNN and Mask-R-CNN, which use
things like region proposals, non-maximum suppression procedure and anchor generation. Moreover, DETR can also be
naturally extended to perform panoptic segmentation, by simply adding a mask head on top of the decoder outputs.
The abstract from the paper is the following:
*We present a new method that views object detection as a direct set prediction problem. Our approach streamlines the
detection pipeline, effectively removing the need for many hand-designed components like a non-maximum suppression
procedure or anchor generation that explicitly encode our prior knowledge about the task. The main ingredients of the
new framework, called DEtection TRansformer or DETR, are a set-based global loss that forces unique predictions via
bipartite matching, and a transformer encoder-decoder architecture. Given a fixed small set of learned object queries,
DETR reasons about the relations of the objects and the global image context to directly output the final set of
predictions in parallel. The new model is conceptually simple and does not require a specialized library, unlike many
other modern detectors. DETR demonstrates accuracy and run-time performance on par with the well-established and
highly-optimized Faster RCNN baseline on the challenging COCO object detection dataset. Moreover, DETR can be easily
generalized to produce panoptic segmentation in a unified manner. We show that it significantly outperforms competitive
baselines.*
This model was contributed by [nielsr](https://huggingface.co/nielsr). The original code can be found [here](https://github.com/facebookresearch/detr).
## How DETR works
Here's a TLDR explaining how [`~transformers.DetrForObjectDetection`] works:
First, an image is sent through a pre-trained convolutional backbone (in the paper, the authors use
ResNet-50/ResNet-101). Let's assume we also add a batch dimension. This means that the input to the backbone is a
tensor of shape `(batch_size, 3, height, width)`, assuming the image has 3 color channels (RGB). The CNN backbone
outputs a new lower-resolution feature map, typically of shape `(batch_size, 2048, height/32, width/32)`. This is
then projected to match the hidden dimension of the Transformer of DETR, which is `256` by default, using a
`nn.Conv2D` layer. So now, we have a tensor of shape `(batch_size, 256, height/32, width/32).` Next, the
feature map is flattened and transposed to obtain a tensor of shape `(batch_size, seq_len, d_model)` =
`(batch_size, width/32*height/32, 256)`. So a difference with NLP models is that the sequence length is actually
longer than usual, but with a smaller `d_model` (which in NLP is typically 768 or higher).
Next, this is sent through the encoder, outputting `encoder_hidden_states` of the same shape (you can consider
these as image features). Next, so-called **object queries** are sent through the decoder. This is a tensor of shape
`(batch_size, num_queries, d_model)`, with `num_queries` typically set to 100 and initialized with zeros.
These input embeddings are learnt positional encodings that the authors refer to as object queries, and similarly to
the encoder, they are added to the input of each attention layer. Each object query will look for a particular object
in the image. The decoder updates these embeddings through multiple self-attention and encoder-decoder attention layers
to output `decoder_hidden_states` of the same shape: `(batch_size, num_queries, d_model)`. Next, two heads
are added on top for object detection: a linear layer for classifying each object query into one of the objects or "no
object", and a MLP to predict bounding boxes for each query.
The model is trained using a **bipartite matching loss**: so what we actually do is compare the predicted classes +
bounding boxes of each of the N = 100 object queries to the ground truth annotations, padded up to the same length N
(so if an image only contains 4 objects, 96 annotations will just have a "no object" as class and "no bounding box" as
bounding box). The [Hungarian matching algorithm](https://en.wikipedia.org/wiki/Hungarian_algorithm) is used to find
an optimal one-to-one mapping of each of the N queries to each of the N annotations. Next, standard cross-entropy (for
the classes) and a linear combination of the L1 and [generalized IoU loss](https://giou.stanford.edu/) (for the
bounding boxes) are used to optimize the parameters of the model.
DETR can be naturally extended to perform panoptic segmentation (which unifies semantic segmentation and instance
segmentation). [`~transformers.DetrForSegmentation`] adds a segmentation mask head on top of
[`~transformers.DetrForObjectDetection`]. The mask head can be trained either jointly, or in a two steps process,
where one first trains a [`~transformers.DetrForObjectDetection`] model to detect bounding boxes around both
"things" (instances) and "stuff" (background things like trees, roads, sky), then freeze all the weights and train only
the mask head for 25 epochs. Experimentally, these two approaches give similar results. Note that predicting boxes is
required for the training to be possible, since the Hungarian matching is computed using distances between boxes.
## Usage tips
- DETR uses so-called **object queries** to detect objects in an image. The number of queries determines the maximum
number of objects that can be detected in a single image, and is set to 100 by default (see parameter
`num_queries` of [`~transformers.DetrConfig`]). Note that it's good to have some slack (in COCO, the
authors used 100, while the maximum number of objects in a COCO image is ~70).
- The decoder of DETR updates the query embeddings in parallel. This is different from language models like GPT-2,
which use autoregressive decoding instead of parallel. Hence, no causal attention mask is used.
- DETR adds position embeddings to the hidden states at each self-attention and cross-attention layer before projecting
to queries and keys. For the position embeddings of the image, one can choose between fixed sinusoidal or learned
absolute position embeddings. By default, the parameter `position_embedding_type` of
[`~transformers.DetrConfig`] is set to `"sine"`.
- During training, the authors of DETR did find it helpful to use auxiliary losses in the decoder, especially to help
the model output the correct number of objects of each class. If you set the parameter `auxiliary_loss` of
[`~transformers.DetrConfig`] to `True`, then prediction feedforward neural networks and Hungarian losses
are added after each decoder layer (with the FFNs sharing parameters).
- If you want to train the model in a distributed environment across multiple nodes, then one should update the
_num_boxes_ variable in the _DetrLoss_ class of _modeling_detr.py_. When training on multiple nodes, this should be
set to the average number of target boxes across all nodes, as can be seen in the original implementation [here](https://github.com/facebookresearch/detr/blob/a54b77800eb8e64e3ad0d8237789fcbf2f8350c5/models/detr.py#L227-L232).
- [`~transformers.DetrForObjectDetection`] and [`~transformers.DetrForSegmentation`] can be initialized with
any convolutional backbone available in the [timm library](https://github.com/rwightman/pytorch-image-models).
Initializing with a MobileNet backbone for example can be done by setting the `backbone` attribute of
[`~transformers.DetrConfig`] to `"tf_mobilenetv3_small_075"`, and then initializing the model with that
config.
- DETR resizes the input images such that the shortest side is at least a certain amount of pixels while the longest is
at most 1333 pixels. At training time, scale augmentation is used such that the shortest side is randomly set to at
least 480 and at most 800 pixels. At inference time, the shortest side is set to 800. One can use
[`~transformers.DetrImageProcessor`] to prepare images (and optional annotations in COCO format) for the
model. Due to this resizing, images in a batch can have different sizes. DETR solves this by padding images up to the
largest size in a batch, and by creating a pixel mask that indicates which pixels are real/which are padding.
Alternatively, one can also define a custom `collate_fn` in order to batch images together, using
[`~transformers.DetrImageProcessor.pad_and_create_pixel_mask`].
- The size of the images will determine the amount of memory being used, and will thus determine the `batch_size`.
It is advised to use a batch size of 2 per GPU. See [this Github thread](https://github.com/facebookresearch/detr/issues/150) for more info.
There are three ways to instantiate a DETR model (depending on what you prefer):
Option 1: Instantiate DETR with pre-trained weights for entire model
```py
>>> from transformers import DetrForObjectDetection
>>> model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")
```
Option 2: Instantiate DETR with randomly initialized weights for Transformer, but pre-trained weights for backbone
```py
>>> from transformers import DetrConfig, DetrForObjectDetection
>>> config = DetrConfig()
>>> model = DetrForObjectDetection(config)
```
Option 3: Instantiate DETR with randomly initialized weights for backbone + Transformer
```py
>>> config = DetrConfig(use_pretrained_backbone=False)
>>> model = DetrForObjectDetection(config)
```
As a summary, consider the following table:
| Task | Object detection | Instance segmentation | Panoptic segmentation |
|------|------------------|-----------------------|-----------------------|
| **Description** | Predicting bounding boxes and class labels around objects in an image | Predicting masks around objects (i.e. instances) in an image | Predicting masks around both objects (i.e. instances) as well as "stuff" (i.e. background things like trees and roads) in an image |
| **Model** | [`~transformers.DetrForObjectDetection`] | [`~transformers.DetrForSegmentation`] | [`~transformers.DetrForSegmentation`] |
| **Example dataset** | COCO detection | COCO detection, COCO panoptic | COCO panoptic | |
| **Format of annotations to provide to** [`~transformers.DetrImageProcessor`] | {'image_id': `int`, 'annotations': `List[Dict]`} each Dict being a COCO object annotation | {'image_id': `int`, 'annotations': `List[Dict]`} (in case of COCO detection) or {'file_name': `str`, 'image_id': `int`, 'segments_info': `List[Dict]`} (in case of COCO panoptic) | {'file_name': `str`, 'image_id': `int`, 'segments_info': `List[Dict]`} and masks_path (path to directory containing PNG files of the masks) |
| **Postprocessing** (i.e. converting the output of the model to Pascal VOC format) | [`~transformers.DetrImageProcessor.post_process`] | [`~transformers.DetrImageProcessor.post_process_segmentation`] | [`~transformers.DetrImageProcessor.post_process_segmentation`], [`~transformers.DetrImageProcessor.post_process_panoptic`] |
| **evaluators** | `CocoEvaluator` with `iou_types="bbox"` | `CocoEvaluator` with `iou_types="bbox"` or `"segm"` | `CocoEvaluator` with `iou_tupes="bbox"` or `"segm"`, `PanopticEvaluator` |
In short, one should prepare the data either in COCO detection or COCO panoptic format, then use
[`~transformers.DetrImageProcessor`] to create `pixel_values`, `pixel_mask` and optional
`labels`, which can then be used to train (or fine-tune) a model. For evaluation, one should first convert the
outputs of the model using one of the postprocessing methods of [`~transformers.DetrImageProcessor`]. These can
be be provided to either `CocoEvaluator` or `PanopticEvaluator`, which allow you to calculate metrics like
mean Average Precision (mAP) and Panoptic Quality (PQ). The latter objects are implemented in the [original repository](https://github.com/facebookresearch/detr). See the [example notebooks](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/DETR) for more info regarding evaluation.
## Resources
A list of official Hugging Face and community (indicated by ð) resources to help you get started with DETR.
<PipelineTag pipeline="object-detection"/>
- All example notebooks illustrating fine-tuning [`DetrForObjectDetection`] and [`DetrForSegmentation`] on a custom dataset an be found [here](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/DETR).
- See also: [Object detection task guide](../tasks/object_detection)
If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
## DetrConfig
[[autodoc]] DetrConfig
## DetrImageProcessor
[[autodoc]] DetrImageProcessor
- preprocess
- post_process_object_detection
- post_process_semantic_segmentation
- post_process_instance_segmentation
- post_process_panoptic_segmentation
## DetrFeatureExtractor
[[autodoc]] DetrFeatureExtractor
- __call__
- post_process_object_detection
- post_process_semantic_segmentation
- post_process_instance_segmentation
- post_process_panoptic_segmentation
## DETR specific outputs
[[autodoc]] models.detr.modeling_detr.DetrModelOutput
[[autodoc]] models.detr.modeling_detr.DetrObjectDetectionOutput
[[autodoc]] models.detr.modeling_detr.DetrSegmentationOutput
## DetrModel
[[autodoc]] DetrModel
- forward
## DetrForObjectDetection
[[autodoc]] DetrForObjectDetection
- forward
## DetrForSegmentation
[[autodoc]] DetrForSegmentation
- forward
| transformers/docs/source/en/model_doc/detr.md/0 | {
"file_path": "transformers/docs/source/en/model_doc/detr.md",
"repo_id": "transformers",
"token_count": 4104
} | 234 |
<!--Copyright 2022 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# ESM
## Overview
This page provides code and pre-trained weights for Transformer protein language models from Meta AI's Fundamental
AI Research Team, providing the state-of-the-art ESMFold and ESM-2, and the previously released ESM-1b and ESM-1v.
Transformer protein language models were introduced in the paper [Biological structure and function emerge from scaling
unsupervised learning to 250 million protein sequences](https://www.pnas.org/content/118/15/e2016239118) by
Alexander Rives, Joshua Meier, Tom Sercu, Siddharth Goyal, Zeming Lin, Jason Liu, Demi Guo, Myle Ott,
C. Lawrence Zitnick, Jerry Ma, and Rob Fergus.
The first version of this paper was [preprinted in 2019](https://www.biorxiv.org/content/10.1101/622803v1?versioned=true).
ESM-2 outperforms all tested single-sequence protein language models across a range of structure prediction tasks,
and enables atomic resolution structure prediction.
It was released with the paper [Language models of protein sequences at the scale of evolution enable accurate
structure prediction](https://doi.org/10.1101/2022.07.20.500902) by Zeming Lin, Halil Akin, Roshan Rao, Brian Hie,
Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido and Alexander Rives.
Also introduced in this paper was ESMFold. It uses an ESM-2 stem with a head that can predict folded protein
structures with state-of-the-art accuracy. Unlike [AlphaFold2](https://www.nature.com/articles/s41586-021-03819-2),
it relies on the token embeddings from the large pre-trained protein language model stem and does not perform a multiple
sequence alignment (MSA) step at inference time, which means that ESMFold checkpoints are fully "standalone" -
they do not require a database of known protein sequences and structures with associated external query tools
to make predictions, and are much faster as a result.
The abstract from
"Biological structure and function emerge from scaling unsupervised learning to 250
million protein sequences" is
*In the field of artificial intelligence, a combination of scale in data and model capacity enabled by unsupervised
learning has led to major advances in representation learning and statistical generation. In the life sciences, the
anticipated growth of sequencing promises unprecedented data on natural sequence diversity. Protein language modeling
at the scale of evolution is a logical step toward predictive and generative artificial intelligence for biology. To
this end, we use unsupervised learning to train a deep contextual language model on 86 billion amino acids across 250
million protein sequences spanning evolutionary diversity. The resulting model contains information about biological
properties in its representations. The representations are learned from sequence data alone. The learned representation
space has a multiscale organization reflecting structure from the level of biochemical properties of amino acids to
remote homology of proteins. Information about secondary and tertiary structure is encoded in the representations and
can be identified by linear projections. Representation learning produces features that generalize across a range of
applications, enabling state-of-the-art supervised prediction of mutational effect and secondary structure and
improving state-of-the-art features for long-range contact prediction.*
The abstract from
"Language models of protein sequences at the scale of evolution enable accurate structure prediction" is
*Large language models have recently been shown to develop emergent capabilities with scale, going beyond
simple pattern matching to perform higher level reasoning and generate lifelike images and text. While
language models trained on protein sequences have been studied at a smaller scale, little is known about
what they learn about biology as they are scaled up. In this work we train models up to 15 billion parameters,
the largest language models of proteins to be evaluated to date. We find that as models are scaled they learn
information enabling the prediction of the three-dimensional structure of a protein at the resolution of
individual atoms. We present ESMFold for high accuracy end-to-end atomic level structure prediction directly
from the individual sequence of a protein. ESMFold has similar accuracy to AlphaFold2 and RoseTTAFold for
sequences with low perplexity that are well understood by the language model. ESMFold inference is an
order of magnitude faster than AlphaFold2, enabling exploration of the structural space of metagenomic
proteins in practical timescales.*
The original code can be found [here](https://github.com/facebookresearch/esm) and was
was developed by the Fundamental AI Research team at Meta AI.
ESM-1b, ESM-1v and ESM-2 were contributed to huggingface by [jasonliu](https://huggingface.co/jasonliu)
and [Matt](https://huggingface.co/Rocketknight1).
ESMFold was contributed to huggingface by [Matt](https://huggingface.co/Rocketknight1) and
[Sylvain](https://huggingface.co/sgugger), with a big thank you to Nikita Smetanin, Roshan Rao and Tom Sercu for their
help throughout the process!
## Usage tips
- ESM models are trained with a masked language modeling (MLM) objective.
- The HuggingFace port of ESMFold uses portions of the [openfold](https://github.com/aqlaboratory/openfold) library. The `openfold` library is licensed under the Apache License 2.0.
## Resources
- [Text classification task guide](../tasks/sequence_classification)
- [Token classification task guide](../tasks/token_classification)
- [Masked language modeling task guide](../tasks/masked_language_modeling)
## EsmConfig
[[autodoc]] EsmConfig
- all
## EsmTokenizer
[[autodoc]] EsmTokenizer
- build_inputs_with_special_tokens
- get_special_tokens_mask
- create_token_type_ids_from_sequences
- save_vocabulary
<frameworkcontent>
<pt>
## EsmModel
[[autodoc]] EsmModel
- forward
## EsmForMaskedLM
[[autodoc]] EsmForMaskedLM
- forward
## EsmForSequenceClassification
[[autodoc]] EsmForSequenceClassification
- forward
## EsmForTokenClassification
[[autodoc]] EsmForTokenClassification
- forward
## EsmForProteinFolding
[[autodoc]] EsmForProteinFolding
- forward
</pt>
<tf>
## TFEsmModel
[[autodoc]] TFEsmModel
- call
## TFEsmForMaskedLM
[[autodoc]] TFEsmForMaskedLM
- call
## TFEsmForSequenceClassification
[[autodoc]] TFEsmForSequenceClassification
- call
## TFEsmForTokenClassification
[[autodoc]] TFEsmForTokenClassification
- call
</tf>
</frameworkcontent>
| transformers/docs/source/en/model_doc/esm.md/0 | {
"file_path": "transformers/docs/source/en/model_doc/esm.md",
"repo_id": "transformers",
"token_count": 1906
} | 235 |
<!--Copyright 2023 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contains specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# MADLAD-400
## Overview
MADLAD-400 models were released in the paper [MADLAD-400: A Multilingual And Document-Level Large Audited Dataset](MADLAD-400: A Multilingual And Document-Level Large Audited Dataset).
The abstract from the paper is the following:
*We introduce MADLAD-400, a manually audited, general domain 3T token monolingual dataset based on CommonCrawl, spanning 419 languages. We discuss
the limitations revealed by self-auditing MADLAD-400, and the role data auditing
had in the dataset creation process. We then train and release a 10.7B-parameter
multilingual machine translation model on 250 billion tokens covering over 450
languages using publicly available data, and find that it is competitive with models
that are significantly larger, and report the results on different domains. In addition, we train a 8B-parameter language model, and assess the results on few-shot
translation. We make the baseline models 1
available to the research community.*
This model was added by [Juarez Bochi](https://huggingface.co/jbochi). The original checkpoints can be found [here](https://github.com/google-research/google-research/tree/master/madlad_400).
This is a machine translation model that supports many low-resource languages, and that is competitive with models that are significantly larger.
One can directly use MADLAD-400 weights without finetuning the model:
```python
>>> from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
>>> model = AutoModelForSeq2SeqLM.from_pretrained("google/madlad400-3b-mt")
>>> tokenizer = AutoTokenizer.from_pretrained("google/madlad400-3b-mt")
>>> inputs = tokenizer("<2pt> I love pizza!", return_tensors="pt")
>>> outputs = model.generate(**inputs)
>>> print(tokenizer.batch_decode(outputs, skip_special_tokens=True))
['Eu amo pizza!']
```
Google has released the following variants:
- [google/madlad400-3b-mt](https://huggingface.co/google/madlad400-3b-mt)
- [google/madlad400-7b-mt](https://huggingface.co/google/madlad400-7b-mt)
- [google/madlad400-7b-mt-bt](https://huggingface.co/google/madlad400-7b-mt-bt)
- [google/madlad400-10b-mt](https://huggingface.co/google/madlad400-10b-mt)
The original checkpoints can be found [here](https://github.com/google-research/google-research/tree/master/madlad_400).
<Tip>
Refer to [T5's documentation page](t5) for all API references, code examples, and notebooks. For more details regarding training and evaluation of the MADLAD-400, refer to the model card.
</Tip>
| transformers/docs/source/en/model_doc/madlad-400.md/0 | {
"file_path": "transformers/docs/source/en/model_doc/madlad-400.md",
"repo_id": "transformers",
"token_count": 930
} | 236 |
<!--Copyright 2020 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# MobileBERT
## Overview
The MobileBERT model was proposed in [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny
Zhou. It's a bidirectional transformer based on the BERT model, which is compressed and accelerated using several
approaches.
The abstract from the paper is the following:
*Natural Language Processing (NLP) has recently achieved great success by using huge pre-trained models with hundreds
of millions of parameters. However, these models suffer from heavy model sizes and high latency such that they cannot
be deployed to resource-limited mobile devices. In this paper, we propose MobileBERT for compressing and accelerating
the popular BERT model. Like the original BERT, MobileBERT is task-agnostic, that is, it can be generically applied to
various downstream NLP tasks via simple fine-tuning. Basically, MobileBERT is a thin version of BERT_LARGE, while
equipped with bottleneck structures and a carefully designed balance between self-attentions and feed-forward networks.
To train MobileBERT, we first train a specially designed teacher model, an inverted-bottleneck incorporated BERT_LARGE
model. Then, we conduct knowledge transfer from this teacher to MobileBERT. Empirical studies show that MobileBERT is
4.3x smaller and 5.5x faster than BERT_BASE while achieving competitive results on well-known benchmarks. On the
natural language inference tasks of GLUE, MobileBERT achieves a GLUEscore o 77.7 (0.6 lower than BERT_BASE), and 62 ms
latency on a Pixel 4 phone. On the SQuAD v1.1/v2.0 question answering task, MobileBERT achieves a dev F1 score of
90.0/79.2 (1.5/2.1 higher than BERT_BASE).*
This model was contributed by [vshampor](https://huggingface.co/vshampor). The original code can be found [here](https://github.com/google-research/google-research/tree/master/mobilebert).
## Usage tips
- MobileBERT is a model with absolute position embeddings so it's usually advised to pad the inputs on the right rather
than the left.
- MobileBERT is similar to BERT and therefore relies on the masked language modeling (MLM) objective. It is therefore
efficient at predicting masked tokens and at NLU in general, but is not optimal for text generation. Models trained
with a causal language modeling (CLM) objective are better in that regard.
## Resources
- [Text classification task guide](../tasks/sequence_classification)
- [Token classification task guide](../tasks/token_classification)
- [Question answering task guide](../tasks/question_answering)
- [Masked language modeling task guide](../tasks/masked_language_modeling)
- [Multiple choice task guide](../tasks/multiple_choice)
## MobileBertConfig
[[autodoc]] MobileBertConfig
## MobileBertTokenizer
[[autodoc]] MobileBertTokenizer
## MobileBertTokenizerFast
[[autodoc]] MobileBertTokenizerFast
## MobileBert specific outputs
[[autodoc]] models.mobilebert.modeling_mobilebert.MobileBertForPreTrainingOutput
[[autodoc]] models.mobilebert.modeling_tf_mobilebert.TFMobileBertForPreTrainingOutput
<frameworkcontent>
<pt>
## MobileBertModel
[[autodoc]] MobileBertModel
- forward
## MobileBertForPreTraining
[[autodoc]] MobileBertForPreTraining
- forward
## MobileBertForMaskedLM
[[autodoc]] MobileBertForMaskedLM
- forward
## MobileBertForNextSentencePrediction
[[autodoc]] MobileBertForNextSentencePrediction
- forward
## MobileBertForSequenceClassification
[[autodoc]] MobileBertForSequenceClassification
- forward
## MobileBertForMultipleChoice
[[autodoc]] MobileBertForMultipleChoice
- forward
## MobileBertForTokenClassification
[[autodoc]] MobileBertForTokenClassification
- forward
## MobileBertForQuestionAnswering
[[autodoc]] MobileBertForQuestionAnswering
- forward
</pt>
<tf>
## TFMobileBertModel
[[autodoc]] TFMobileBertModel
- call
## TFMobileBertForPreTraining
[[autodoc]] TFMobileBertForPreTraining
- call
## TFMobileBertForMaskedLM
[[autodoc]] TFMobileBertForMaskedLM
- call
## TFMobileBertForNextSentencePrediction
[[autodoc]] TFMobileBertForNextSentencePrediction
- call
## TFMobileBertForSequenceClassification
[[autodoc]] TFMobileBertForSequenceClassification
- call
## TFMobileBertForMultipleChoice
[[autodoc]] TFMobileBertForMultipleChoice
- call
## TFMobileBertForTokenClassification
[[autodoc]] TFMobileBertForTokenClassification
- call
## TFMobileBertForQuestionAnswering
[[autodoc]] TFMobileBertForQuestionAnswering
- call
</tf>
</frameworkcontent>
| transformers/docs/source/en/model_doc/mobilebert.md/0 | {
"file_path": "transformers/docs/source/en/model_doc/mobilebert.md",
"repo_id": "transformers",
"token_count": 1548
} | 237 |
<!--Copyright 2022 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# Nyströmformer
## Overview
The Nyströmformer model was proposed in [*Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention*](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn
Fung, Yin Li, and Vikas Singh.
The abstract from the paper is the following:
*Transformers have emerged as a powerful tool for a broad range of natural language processing tasks. A key component
that drives the impressive performance of Transformers is the self-attention mechanism that encodes the influence or
dependence of other tokens on each specific token. While beneficial, the quadratic complexity of self-attention on the
input sequence length has limited its application to longer sequences -- a topic being actively studied in the
community. To address this limitation, we propose Nyströmformer -- a model that exhibits favorable scalability as a
function of sequence length. Our idea is based on adapting the Nyström method to approximate standard self-attention
with O(n) complexity. The scalability of Nyströmformer enables application to longer sequences with thousands of
tokens. We perform evaluations on multiple downstream tasks on the GLUE benchmark and IMDB reviews with standard
sequence length, and find that our Nyströmformer performs comparably, or in a few cases, even slightly better, than
standard self-attention. On longer sequence tasks in the Long Range Arena (LRA) benchmark, Nyströmformer performs
favorably relative to other efficient self-attention methods. Our code is available at this https URL.*
This model was contributed by [novice03](https://huggingface.co/novice03). The original code can be found [here](https://github.com/mlpen/Nystromformer).
## Resources
- [Text classification task guide](../tasks/sequence_classification)
- [Token classification task guide](../tasks/token_classification)
- [Question answering task guide](../tasks/question_answering)
- [Masked language modeling task guide](../tasks/masked_language_modeling)
- [Multiple choice task guide](../tasks/multiple_choice)
## NystromformerConfig
[[autodoc]] NystromformerConfig
## NystromformerModel
[[autodoc]] NystromformerModel
- forward
## NystromformerForMaskedLM
[[autodoc]] NystromformerForMaskedLM
- forward
## NystromformerForSequenceClassification
[[autodoc]] NystromformerForSequenceClassification
- forward
## NystromformerForMultipleChoice
[[autodoc]] NystromformerForMultipleChoice
- forward
## NystromformerForTokenClassification
[[autodoc]] NystromformerForTokenClassification
- forward
## NystromformerForQuestionAnswering
[[autodoc]] NystromformerForQuestionAnswering
- forward
| transformers/docs/source/en/model_doc/nystromformer.md/0 | {
"file_path": "transformers/docs/source/en/model_doc/nystromformer.md",
"repo_id": "transformers",
"token_count": 907
} | 238 |
<!--Copyright 2022 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# PLBart
## Overview
The PLBART model was proposed in [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
This is a BART-like model which can be used to perform code-summarization, code-generation, and code-translation tasks. The pre-trained model `plbart-base` has been trained using multilingual denoising task
on Java, Python and English.
According to the abstract
*Code summarization and generation empower conversion between programming language (PL) and natural language (NL),
while code translation avails the migration of legacy code from one PL to another. This paper introduces PLBART,
a sequence-to-sequence model capable of performing a broad spectrum of program and language understanding and generation tasks.
PLBART is pre-trained on an extensive collection of Java and Python functions and associated NL text via denoising autoencoding.
Experiments on code summarization in the English language, code generation, and code translation in seven programming languages
show that PLBART outperforms or rivals state-of-the-art models. Moreover, experiments on discriminative tasks, e.g., program
repair, clone detection, and vulnerable code detection, demonstrate PLBART's effectiveness in program understanding.
Furthermore, analysis reveals that PLBART learns program syntax, style (e.g., identifier naming convention), logical flow
(e.g., if block inside an else block is equivalent to else if block) that are crucial to program semantics and thus excels
even with limited annotations.*
This model was contributed by [gchhablani](https://huggingface.co/gchhablani). The Authors' code can be found [here](https://github.com/wasiahmad/PLBART).
## Usage examples
PLBart is a multilingual encoder-decoder (sequence-to-sequence) model primarily intended for code-to-text, text-to-code, code-to-code tasks. As the
model is multilingual it expects the sequences in a different format. A special language id token is added in both the
source and target text. The source text format is `X [eos, src_lang_code]` where `X` is the source text. The
target text format is `[tgt_lang_code] X [eos]`. `bos` is never used.
However, for fine-tuning, in some cases no language token is provided in cases where a single language is used. Please refer to [the paper](https://arxiv.org/abs/2103.06333) to learn more about this.
In cases where the language code is needed, the regular [`~PLBartTokenizer.__call__`] will encode source text format
when you pass texts as the first argument or with the keyword argument `text`, and will encode target text format if
it's passed with the `text_target` keyword argument.
### Supervised training
```python
>>> from transformers import PLBartForConditionalGeneration, PLBartTokenizer
>>> tokenizer = PLBartTokenizer.from_pretrained("uclanlp/plbart-base", src_lang="en_XX", tgt_lang="python")
>>> example_python_phrase = "def maximum(a,b,c):NEW_LINE_INDENTreturn max([a,b,c])"
>>> expected_translation_english = "Returns the maximum value of a b c."
>>> inputs = tokenizer(example_python_phrase, text_target=expected_translation_english, return_tensors="pt")
>>> model(**inputs)
```
### Generation
While generating the target text set the `decoder_start_token_id` to the target language id. The following
example shows how to translate Python to English using the `uclanlp/plbart-python-en_XX` model.
```python
>>> from transformers import PLBartForConditionalGeneration, PLBartTokenizer
>>> tokenizer = PLBartTokenizer.from_pretrained("uclanlp/plbart-python-en_XX", src_lang="python", tgt_lang="en_XX")
>>> example_python_phrase = "def maximum(a,b,c):NEW_LINE_INDENTreturn max([a,b,c])"
>>> inputs = tokenizer(example_python_phrase, return_tensors="pt")
>>> model = PLBartForConditionalGeneration.from_pretrained("uclanlp/plbart-python-en_XX")
>>> translated_tokens = model.generate(**inputs, decoder_start_token_id=tokenizer.lang_code_to_id["en_XX"])
>>> tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)[0]
"Returns the maximum value of a b c."
```
## Resources
- [Text classification task guide](../tasks/sequence_classification)
- [Causal language modeling task guide](../tasks/language_modeling)
- [Translation task guide](../tasks/translation)
- [Summarization task guide](../tasks/summarization)
## PLBartConfig
[[autodoc]] PLBartConfig
## PLBartTokenizer
[[autodoc]] PLBartTokenizer
- build_inputs_with_special_tokens
## PLBartModel
[[autodoc]] PLBartModel
- forward
## PLBartForConditionalGeneration
[[autodoc]] PLBartForConditionalGeneration
- forward
## PLBartForSequenceClassification
[[autodoc]] PLBartForSequenceClassification
- forward
## PLBartForCausalLM
[[autodoc]] PLBartForCausalLM
- forward | transformers/docs/source/en/model_doc/plbart.md/0 | {
"file_path": "transformers/docs/source/en/model_doc/plbart.md",
"repo_id": "transformers",
"token_count": 1586
} | 239 |
<!--Copyright 2022 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# RoCBert
## Overview
The RoCBert model was proposed in [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
It's a pretrained Chinese language model that is robust under various forms of adversarial attacks.
The abstract from the paper is the following:
*Large-scale pretrained language models have achieved SOTA results on NLP tasks. However, they have been shown
vulnerable to adversarial attacks especially for logographic languages like Chinese. In this work, we propose
ROCBERT: a pretrained Chinese Bert that is robust to various forms of adversarial attacks like word perturbation,
synonyms, typos, etc. It is pretrained with the contrastive learning objective which maximizes the label consistency
under different synthesized adversarial examples. The model takes as input multimodal information including the
semantic, phonetic and visual features. We show all these features are important to the model robustness since the
attack can be performed in all the three forms. Across 5 Chinese NLU tasks, ROCBERT outperforms strong baselines under
three blackbox adversarial algorithms without sacrificing the performance on clean testset. It also performs the best
in the toxic content detection task under human-made attacks.*
This model was contributed by [weiweishi](https://huggingface.co/weiweishi).
## Resources
- [Text classification task guide](../tasks/sequence_classification)
- [Token classification task guide](../tasks/token_classification)
- [Question answering task guide](../tasks/question_answering)
- [Causal language modeling task guide](../tasks/language_modeling)
- [Masked language modeling task guide](../tasks/masked_language_modeling)
- [Multiple choice task guide](../tasks/multiple_choice)
## RoCBertConfig
[[autodoc]] RoCBertConfig
- all
## RoCBertTokenizer
[[autodoc]] RoCBertTokenizer
- build_inputs_with_special_tokens
- get_special_tokens_mask
- create_token_type_ids_from_sequences
- save_vocabulary
## RoCBertModel
[[autodoc]] RoCBertModel
- forward
## RoCBertForPreTraining
[[autodoc]] RoCBertForPreTraining
- forward
## RoCBertForCausalLM
[[autodoc]] RoCBertForCausalLM
- forward
## RoCBertForMaskedLM
[[autodoc]] RoCBertForMaskedLM
- forward
## RoCBertForSequenceClassification
[[autodoc]] transformers.RoCBertForSequenceClassification
- forward
## RoCBertForMultipleChoice
[[autodoc]] transformers.RoCBertForMultipleChoice
- forward
## RoCBertForTokenClassification
[[autodoc]] transformers.RoCBertForTokenClassification
- forward
## RoCBertForQuestionAnswering
[[autodoc]] RoCBertForQuestionAnswering
- forward
| transformers/docs/source/en/model_doc/roc_bert.md/0 | {
"file_path": "transformers/docs/source/en/model_doc/roc_bert.md",
"repo_id": "transformers",
"token_count": 999
} | 240 |
<!--Copyright 2022 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# ViTMAE
## Overview
The ViTMAE model was proposed in [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377v2) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li,
Piotr Dollár, Ross Girshick. The paper shows that, by pre-training a Vision Transformer (ViT) to reconstruct pixel values for masked patches, one can get results after
fine-tuning that outperform supervised pre-training.
The abstract from the paper is the following:
*This paper shows that masked autoencoders (MAE) are scalable self-supervised learners for computer vision. Our MAE approach is simple: we mask random patches of the
input image and reconstruct the missing pixels. It is based on two core designs. First, we develop an asymmetric encoder-decoder architecture, with an encoder that operates
only on the visible subset of patches (without mask tokens), along with a lightweight decoder that reconstructs the original image from the latent representation and mask
tokens. Second, we find that masking a high proportion of the input image, e.g., 75%, yields a nontrivial and meaningful self-supervisory task. Coupling these two designs
enables us to train large models efficiently and effectively: we accelerate training (by 3x or more) and improve accuracy. Our scalable approach allows for learning high-capacity
models that generalize well: e.g., a vanilla ViT-Huge model achieves the best accuracy (87.8%) among methods that use only ImageNet-1K data. Transfer performance in downstream
tasks outperforms supervised pre-training and shows promising scaling behavior.*
<img src="https://user-images.githubusercontent.com/11435359/146857310-f258c86c-fde6-48e8-9cee-badd2b21bd2c.png"
alt="drawing" width="600"/>
<small> MAE architecture. Taken from the <a href="https://arxiv.org/abs/2111.06377">original paper.</a> </small>
This model was contributed by [nielsr](https://huggingface.co/nielsr). TensorFlow version of the model was contributed by [sayakpaul](https://github.com/sayakpaul) and
[ariG23498](https://github.com/ariG23498) (equal contribution). The original code can be found [here](https://github.com/facebookresearch/mae).
## Usage tips
- MAE (masked auto encoding) is a method for self-supervised pre-training of Vision Transformers (ViTs). The pre-training objective is relatively simple:
by masking a large portion (75%) of the image patches, the model must reconstruct raw pixel values. One can use [`ViTMAEForPreTraining`] for this purpose.
- After pre-training, one "throws away" the decoder used to reconstruct pixels, and one uses the encoder for fine-tuning/linear probing. This means that after
fine-tuning, one can directly plug in the weights into a [`ViTForImageClassification`].
- One can use [`ViTImageProcessor`] to prepare images for the model. See the code examples for more info.
- Note that the encoder of MAE is only used to encode the visual patches. The encoded patches are then concatenated with mask tokens, which the decoder (which also
consists of Transformer blocks) takes as input. Each mask token is a shared, learned vector that indicates the presence of a missing patch to be predicted. Fixed
sin/cos position embeddings are added both to the input of the encoder and the decoder.
- For a visual understanding of how MAEs work you can check out this [post](https://keras.io/examples/vision/masked_image_modeling/).
## Resources
A list of official Hugging Face and community (indicated by ð) resources to help you get started with ViTMAE.
- [`ViTMAEForPreTraining`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-pretraining), allowing you to pre-train the model from scratch/further pre-train the model on custom data.
- A notebook that illustrates how to visualize reconstructed pixel values with [`ViTMAEForPreTraining`] can be found [here](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/ViTMAE/ViT_MAE_visualization_demo.ipynb).
If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
## ViTMAEConfig
[[autodoc]] ViTMAEConfig
<frameworkcontent>
<pt>
## ViTMAEModel
[[autodoc]] ViTMAEModel
- forward
## ViTMAEForPreTraining
[[autodoc]] transformers.ViTMAEForPreTraining
- forward
</pt>
<tf>
## TFViTMAEModel
[[autodoc]] TFViTMAEModel
- call
## TFViTMAEForPreTraining
[[autodoc]] transformers.TFViTMAEForPreTraining
- call
</tf>
</frameworkcontent>
| transformers/docs/source/en/model_doc/vit_mae.md/0 | {
"file_path": "transformers/docs/source/en/model_doc/vit_mae.md",
"repo_id": "transformers",
"token_count": 1492
} | 241 |
<!--Copyright 2020 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# XLM-RoBERTa
<div class="flex flex-wrap space-x-1">
<a href="https://huggingface.co/models?filter=xlm-roberta">
<img alt="Models" src="https://img.shields.io/badge/All_model_pages-xlm--roberta-blueviolet">
</a>
<a href="https://huggingface.co/spaces/docs-demos/xlm-roberta-base">
<img alt="Spaces" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue">
</a>
</div>
## Overview
The XLM-RoBERTa model was proposed in [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume
Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov. It is based on Facebook's
RoBERTa model released in 2019. It is a large multi-lingual language model, trained on 2.5TB of filtered CommonCrawl
data.
The abstract from the paper is the following:
*This paper shows that pretraining multilingual language models at scale leads to significant performance gains for a
wide range of cross-lingual transfer tasks. We train a Transformer-based masked language model on one hundred
languages, using more than two terabytes of filtered CommonCrawl data. Our model, dubbed XLM-R, significantly
outperforms multilingual BERT (mBERT) on a variety of cross-lingual benchmarks, including +13.8% average accuracy on
XNLI, +12.3% average F1 score on MLQA, and +2.1% average F1 score on NER. XLM-R performs particularly well on
low-resource languages, improving 11.8% in XNLI accuracy for Swahili and 9.2% for Urdu over the previous XLM model. We
also present a detailed empirical evaluation of the key factors that are required to achieve these gains, including the
trade-offs between (1) positive transfer and capacity dilution and (2) the performance of high and low resource
languages at scale. Finally, we show, for the first time, the possibility of multilingual modeling without sacrificing
per-language performance; XLM-Ris very competitive with strong monolingual models on the GLUE and XNLI benchmarks. We
will make XLM-R code, data, and models publicly available.*
This model was contributed by [stefan-it](https://huggingface.co/stefan-it). The original code can be found [here](https://github.com/pytorch/fairseq/tree/master/examples/xlmr).
## Usage tips
- XLM-RoBERTa is a multilingual model trained on 100 different languages. Unlike some XLM multilingual models, it does
not require `lang` tensors to understand which language is used, and should be able to determine the correct
language from the input ids.
- Uses RoBERTa tricks on the XLM approach, but does not use the translation language modeling objective. It only uses masked language modeling on sentences coming from one language.
## Resources
A list of official Hugging Face and community (indicated by ð) resources to help you get started with XLM-RoBERTa. If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
<PipelineTag pipeline="text-classification"/>
- A blog post on how to [finetune XLM RoBERTa for multiclass classification with Habana Gaudi on AWS](https://www.philschmid.de/habana-distributed-training)
- [`XLMRobertaForSequenceClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/text-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/text_classification.ipynb).
- [`TFXLMRobertaForSequenceClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/text-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/text_classification-tf.ipynb).
- [`FlaxXLMRobertaForSequenceClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/text-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/text_classification_flax.ipynb).
- [Text classification](https://huggingface.co/docs/transformers/tasks/sequence_classification) chapter of the ð€ Hugging Face Task Guides.
- [Text classification task guide](../tasks/sequence_classification)
<PipelineTag pipeline="token-classification"/>
- [`XLMRobertaForTokenClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/token-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/token_classification.ipynb).
- [`TFXLMRobertaForTokenClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/token-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/token_classification-tf.ipynb).
- [`FlaxXLMRobertaForTokenClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/token-classification).
- [Token classification](https://huggingface.co/course/chapter7/2?fw=pt) chapter of the ð€ Hugging Face Course.
- [Token classification task guide](../tasks/token_classification)
<PipelineTag pipeline="text-generation"/>
- [`XLMRobertaForCausalLM`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/language-modeling) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling.ipynb).
- [Causal language modeling](https://huggingface.co/docs/transformers/tasks/language_modeling) chapter of the ð€ Hugging Face Task Guides.
- [Causal language modeling task guide](../tasks/language_modeling)
<PipelineTag pipeline="fill-mask"/>
- [`XLMRobertaForMaskedLM`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/language-modeling#robertabertdistilbert-and-masked-language-modeling) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling.ipynb).
- [`TFXLMRobertaForMaskedLM`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/language-modeling#run_mlmpy) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling-tf.ipynb).
- [`FlaxXLMRobertaForMaskedLM`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/language-modeling#masked-language-modeling) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/masked_language_modeling_flax.ipynb).
- [Masked language modeling](https://huggingface.co/course/chapter7/3?fw=pt) chapter of the ð€ Hugging Face Course.
- [Masked language modeling](../tasks/masked_language_modeling)
<PipelineTag pipeline="question-answering"/>
- [`XLMRobertaForQuestionAnswering`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/question-answering) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/question_answering.ipynb).
- [`TFXLMRobertaForQuestionAnswering`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/question-answering) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/question_answering-tf.ipynb).
- [`FlaxXLMRobertaForQuestionAnswering`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/question-answering).
- [Question answering](https://huggingface.co/course/chapter7/7?fw=pt) chapter of the ð€ Hugging Face Course.
- [Question answering task guide](../tasks/question_answering)
**Multiple choice**
- [`XLMRobertaForMultipleChoice`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/multiple-choice) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/multiple_choice.ipynb).
- [`TFXLMRobertaForMultipleChoice`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/multiple-choice) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/multiple_choice-tf.ipynb).
- [Multiple choice task guide](../tasks/multiple_choice)
ð Deploy
- A blog post on how to [Deploy Serverless XLM RoBERTa on AWS Lambda](https://www.philschmid.de/multilingual-serverless-xlm-roberta-with-huggingface).
<Tip>
This implementation is the same as RoBERTa. Refer to the [documentation of RoBERTa](roberta) for usage examples as well as the information relative to the inputs and outputs.
</Tip>
## XLMRobertaConfig
[[autodoc]] XLMRobertaConfig
## XLMRobertaTokenizer
[[autodoc]] XLMRobertaTokenizer
- build_inputs_with_special_tokens
- get_special_tokens_mask
- create_token_type_ids_from_sequences
- save_vocabulary
## XLMRobertaTokenizerFast
[[autodoc]] XLMRobertaTokenizerFast
<frameworkcontent>
<pt>
## XLMRobertaModel
[[autodoc]] XLMRobertaModel
- forward
## XLMRobertaForCausalLM
[[autodoc]] XLMRobertaForCausalLM
- forward
## XLMRobertaForMaskedLM
[[autodoc]] XLMRobertaForMaskedLM
- forward
## XLMRobertaForSequenceClassification
[[autodoc]] XLMRobertaForSequenceClassification
- forward
## XLMRobertaForMultipleChoice
[[autodoc]] XLMRobertaForMultipleChoice
- forward
## XLMRobertaForTokenClassification
[[autodoc]] XLMRobertaForTokenClassification
- forward
## XLMRobertaForQuestionAnswering
[[autodoc]] XLMRobertaForQuestionAnswering
- forward
</pt>
<tf>
## TFXLMRobertaModel
[[autodoc]] TFXLMRobertaModel
- call
## TFXLMRobertaForCausalLM
[[autodoc]] TFXLMRobertaForCausalLM
- call
## TFXLMRobertaForMaskedLM
[[autodoc]] TFXLMRobertaForMaskedLM
- call
## TFXLMRobertaForSequenceClassification
[[autodoc]] TFXLMRobertaForSequenceClassification
- call
## TFXLMRobertaForMultipleChoice
[[autodoc]] TFXLMRobertaForMultipleChoice
- call
## TFXLMRobertaForTokenClassification
[[autodoc]] TFXLMRobertaForTokenClassification
- call
## TFXLMRobertaForQuestionAnswering
[[autodoc]] TFXLMRobertaForQuestionAnswering
- call
</tf>
<jax>
## FlaxXLMRobertaModel
[[autodoc]] FlaxXLMRobertaModel
- __call__
## FlaxXLMRobertaForCausalLM
[[autodoc]] FlaxXLMRobertaForCausalLM
- __call__
## FlaxXLMRobertaForMaskedLM
[[autodoc]] FlaxXLMRobertaForMaskedLM
- __call__
## FlaxXLMRobertaForSequenceClassification
[[autodoc]] FlaxXLMRobertaForSequenceClassification
- __call__
## FlaxXLMRobertaForMultipleChoice
[[autodoc]] FlaxXLMRobertaForMultipleChoice
- __call__
## FlaxXLMRobertaForTokenClassification
[[autodoc]] FlaxXLMRobertaForTokenClassification
- __call__
## FlaxXLMRobertaForQuestionAnswering
[[autodoc]] FlaxXLMRobertaForQuestionAnswering
- __call__
</jax>
</frameworkcontent> | transformers/docs/source/en/model_doc/xlm-roberta.md/0 | {
"file_path": "transformers/docs/source/en/model_doc/xlm-roberta.md",
"repo_id": "transformers",
"token_count": 3907
} | 242 |
<!---
Copyright 2022 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# Custom hardware for training
The hardware you use to run model training and inference can have a big effect on performance. For a deep dive into GPUs make sure to check out Tim Dettmer's excellent [blog post](https://timdettmers.com/2020/09/07/which-gpu-for-deep-learning/).
Let's have a look at some practical advice for GPU setups.
## GPU
When you train bigger models you have essentially three options:
- bigger GPUs
- more GPUs
- more CPU and NVMe (offloaded to by [DeepSpeed-Infinity](main_classes/deepspeed#nvme-support))
Let's start at the case where you have a single GPU.
### Power and Cooling
If you bought an expensive high end GPU make sure you give it the correct power and sufficient cooling.
**Power**:
Some high end consumer GPU cards have 2 and sometimes 3 PCI-E 8-Pin power sockets. Make sure you have as many independent 12V PCI-E 8-Pin cables plugged into the card as there are sockets. Do not use the 2 splits at one end of the same cable (also known as pigtail cable). That is if you have 2 sockets on the GPU, you want 2 PCI-E 8-Pin cables going from your PSU to the card and not one that has 2 PCI-E 8-Pin connectors at the end! You won't get the full performance out of your card otherwise.
Each PCI-E 8-Pin power cable needs to be plugged into a 12V rail on the PSU side and can supply up to 150W of power.
Some other cards may use a PCI-E 12-Pin connectors, and these can deliver up to 500-600W of power.
Low end cards may use 6-Pin connectors, which supply up to 75W of power.
Additionally you want the high-end PSU that has stable voltage. Some lower quality ones may not give the card the stable voltage it needs to function at its peak.
And of course the PSU needs to have enough unused Watts to power the card.
**Cooling**:
When a GPU gets overheated it will start throttling down and will not deliver full performance and it can even shutdown if it gets too hot.
It's hard to tell the exact best temperature to strive for when a GPU is heavily loaded, but probably anything under +80C is good, but lower is better - perhaps 70-75C is an excellent range to be in. The throttling down is likely to start at around 84-90C. But other than throttling performance a prolonged very high temperature is likely to reduce the lifespan of a GPU.
Next let's have a look at one of the most important aspects when having multiple GPUs: connectivity.
### Multi-GPU Connectivity
If you use multiple GPUs the way cards are inter-connected can have a huge impact on the total training time. If the GPUs are on the same physical node, you can run:
```
nvidia-smi topo -m
```
and it will tell you how the GPUs are inter-connected. On a machine with dual-GPU and which are connected with NVLink, you will most likely see something like:
```
GPU0 GPU1 CPU Affinity NUMA Affinity
GPU0 X NV2 0-23 N/A
GPU1 NV2 X 0-23 N/A
```
on a different machine w/o NVLink we may see:
```
GPU0 GPU1 CPU Affinity NUMA Affinity
GPU0 X PHB 0-11 N/A
GPU1 PHB X 0-11 N/A
```
The report includes this legend:
```
X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks
```
So the first report `NV2` tells us the GPUs are interconnected with 2 NVLinks, and the second report `PHB` we have a typical consumer-level PCIe+Bridge setup.
Check what type of connectivity you have on your setup. Some of these will make the communication between cards faster (e.g. NVLink), others slower (e.g. PHB).
Depending on the type of scalability solution used, the connectivity speed could have a major or a minor impact. If the GPUs need to sync rarely, as in DDP, the impact of a slower connection will be less significant. If the GPUs need to send messages to each other often, as in ZeRO-DP, then faster connectivity becomes super important to achieve faster training.
#### NVlink
[NVLink](https://en.wikipedia.org/wiki/NVLink) is a wire-based serial multi-lane near-range communications link developed by Nvidia.
Each new generation provides a faster bandwidth, e.g. here is a quote from [Nvidia Ampere GA102 GPU Architecture](https://www.nvidia.com/content/dam/en-zz/Solutions/geforce/ampere/pdf/NVIDIA-ampere-GA102-GPU-Architecture-Whitepaper-V1.pdf):
> Third-Generation NVLink®
> GA102 GPUs utilize NVIDIAâs third-generation NVLink interface, which includes four x4 links,
> with each link providing 14.0625 GB/sec bandwidth in each direction between two GPUs. Four
> links provide 56.25 GB/sec bandwidth in each direction, and 112.5 GB/sec total bandwidth
> between two GPUs. Two RTX 3090 GPUs can be connected together for SLI using NVLink.
> (Note that 3-Way and 4-Way SLI configurations are not supported.)
So the higher `X` you get in the report of `NVX` in the output of `nvidia-smi topo -m` the better. The generation will depend on your GPU architecture.
Let's compare the execution of a gpt2 language model training over a small sample of wikitext.
The results are:
| NVlink | Time |
| ----- | ---: |
| Y | 101s |
| N | 131s |
You can see that NVLink completes the training ~23% faster. In the second benchmark we use `NCCL_P2P_DISABLE=1` to tell the GPUs not to use NVLink.
Here is the full benchmark code and outputs:
```bash
# DDP w/ NVLink
rm -r /tmp/test-clm; CUDA_VISIBLE_DEVICES=0,1 torchrun \
--nproc_per_node 2 examples/pytorch/language-modeling/run_clm.py --model_name_or_path gpt2 \
--dataset_name wikitext --dataset_config_name wikitext-2-raw-v1 --do_train \
--output_dir /tmp/test-clm --per_device_train_batch_size 4 --max_steps 200
{'train_runtime': 101.9003, 'train_samples_per_second': 1.963, 'epoch': 0.69}
# DDP w/o NVLink
rm -r /tmp/test-clm; CUDA_VISIBLE_DEVICES=0,1 NCCL_P2P_DISABLE=1 torchrun \
--nproc_per_node 2 examples/pytorch/language-modeling/run_clm.py --model_name_or_path gpt2 \
--dataset_name wikitext --dataset_config_name wikitext-2-raw-v1 --do_train
--output_dir /tmp/test-clm --per_device_train_batch_size 4 --max_steps 200
{'train_runtime': 131.4367, 'train_samples_per_second': 1.522, 'epoch': 0.69}
```
Hardware: 2x TITAN RTX 24GB each + NVlink with 2 NVLinks (`NV2` in `nvidia-smi topo -m`)
Software: `pytorch-1.8-to-be` + `cuda-11.0` / `transformers==4.3.0.dev0`
| transformers/docs/source/en/perf_hardware.md/0 | {
"file_path": "transformers/docs/source/en/perf_hardware.md",
"repo_id": "transformers",
"token_count": 2301
} | 243 |
<!--Copyright 2023 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# Preprocess
[[open-in-colab]]
Before you can train a model on a dataset, it needs to be preprocessed into the expected model input format. Whether your data is text, images, or audio, they need to be converted and assembled into batches of tensors. ð€ Transformers provides a set of preprocessing classes to help prepare your data for the model. In this tutorial, you'll learn that for:
* Text, use a [Tokenizer](./main_classes/tokenizer) to convert text into a sequence of tokens, create a numerical representation of the tokens, and assemble them into tensors.
* Speech and audio, use a [Feature extractor](./main_classes/feature_extractor) to extract sequential features from audio waveforms and convert them into tensors.
* Image inputs use a [ImageProcessor](./main_classes/image_processor) to convert images into tensors.
* Multimodal inputs, use a [Processor](./main_classes/processors) to combine a tokenizer and a feature extractor or image processor.
<Tip>
`AutoProcessor` **always** works and automatically chooses the correct class for the model you're using, whether you're using a tokenizer, image processor, feature extractor or processor.
</Tip>
Before you begin, install ð€ Datasets so you can load some datasets to experiment with:
```bash
pip install datasets
```
## Natural Language Processing
<Youtube id="Yffk5aydLzg"/>
The main tool for preprocessing textual data is a [tokenizer](main_classes/tokenizer). A tokenizer splits text into *tokens* according to a set of rules. The tokens are converted into numbers and then tensors, which become the model inputs. Any additional inputs required by the model are added by the tokenizer.
<Tip>
If you plan on using a pretrained model, it's important to use the associated pretrained tokenizer. This ensures the text is split the same way as the pretraining corpus, and uses the same corresponding tokens-to-index (usually referred to as the *vocab*) during pretraining.
</Tip>
Get started by loading a pretrained tokenizer with the [`AutoTokenizer.from_pretrained`] method. This downloads the *vocab* a model was pretrained with:
```py
>>> from transformers import AutoTokenizer
>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
```
Then pass your text to the tokenizer:
```py
>>> encoded_input = tokenizer("Do not meddle in the affairs of wizards, for they are subtle and quick to anger.")
>>> print(encoded_input)
{'input_ids': [101, 2079, 2025, 19960, 10362, 1999, 1996, 3821, 1997, 16657, 1010, 2005, 2027, 2024, 11259, 1998, 4248, 2000, 4963, 1012, 102],
'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]}
```
The tokenizer returns a dictionary with three important items:
* [input_ids](glossary#input-ids) are the indices corresponding to each token in the sentence.
* [attention_mask](glossary#attention-mask) indicates whether a token should be attended to or not.
* [token_type_ids](glossary#token-type-ids) identifies which sequence a token belongs to when there is more than one sequence.
Return your input by decoding the `input_ids`:
```py
>>> tokenizer.decode(encoded_input["input_ids"])
'[CLS] Do not meddle in the affairs of wizards, for they are subtle and quick to anger. [SEP]'
```
As you can see, the tokenizer added two special tokens - `CLS` and `SEP` (classifier and separator) - to the sentence. Not all models need
special tokens, but if they do, the tokenizer automatically adds them for you.
If there are several sentences you want to preprocess, pass them as a list to the tokenizer:
```py
>>> batch_sentences = [
... "But what about second breakfast?",
... "Don't think he knows about second breakfast, Pip.",
... "What about elevensies?",
... ]
>>> encoded_inputs = tokenizer(batch_sentences)
>>> print(encoded_inputs)
{'input_ids': [[101, 1252, 1184, 1164, 1248, 6462, 136, 102],
[101, 1790, 112, 189, 1341, 1119, 3520, 1164, 1248, 6462, 117, 21902, 1643, 119, 102],
[101, 1327, 1164, 5450, 23434, 136, 102]],
'token_type_ids': [[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0]],
'attention_mask': [[1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1]]}
```
### Pad
Sentences aren't always the same length which can be an issue because tensors, the model inputs, need to have a uniform shape. Padding is a strategy for ensuring tensors are rectangular by adding a special *padding token* to shorter sentences.
Set the `padding` parameter to `True` to pad the shorter sequences in the batch to match the longest sequence:
```py
>>> batch_sentences = [
... "But what about second breakfast?",
... "Don't think he knows about second breakfast, Pip.",
... "What about elevensies?",
... ]
>>> encoded_input = tokenizer(batch_sentences, padding=True)
>>> print(encoded_input)
{'input_ids': [[101, 1252, 1184, 1164, 1248, 6462, 136, 102, 0, 0, 0, 0, 0, 0, 0],
[101, 1790, 112, 189, 1341, 1119, 3520, 1164, 1248, 6462, 117, 21902, 1643, 119, 102],
[101, 1327, 1164, 5450, 23434, 136, 102, 0, 0, 0, 0, 0, 0, 0, 0]],
'token_type_ids': [[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],
'attention_mask': [[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0]]}
```
The first and third sentences are now padded with `0`'s because they are shorter.
### Truncation
On the other end of the spectrum, sometimes a sequence may be too long for a model to handle. In this case, you'll need to truncate the sequence to a shorter length.
Set the `truncation` parameter to `True` to truncate a sequence to the maximum length accepted by the model:
```py
>>> batch_sentences = [
... "But what about second breakfast?",
... "Don't think he knows about second breakfast, Pip.",
... "What about elevensies?",
... ]
>>> encoded_input = tokenizer(batch_sentences, padding=True, truncation=True)
>>> print(encoded_input)
{'input_ids': [[101, 1252, 1184, 1164, 1248, 6462, 136, 102, 0, 0, 0, 0, 0, 0, 0],
[101, 1790, 112, 189, 1341, 1119, 3520, 1164, 1248, 6462, 117, 21902, 1643, 119, 102],
[101, 1327, 1164, 5450, 23434, 136, 102, 0, 0, 0, 0, 0, 0, 0, 0]],
'token_type_ids': [[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],
'attention_mask': [[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0]]}
```
<Tip>
Check out the [Padding and truncation](./pad_truncation) concept guide to learn more different padding and truncation arguments.
</Tip>
### Build tensors
Finally, you want the tokenizer to return the actual tensors that get fed to the model.
Set the `return_tensors` parameter to either `pt` for PyTorch, or `tf` for TensorFlow:
<frameworkcontent>
<pt>
```py
>>> batch_sentences = [
... "But what about second breakfast?",
... "Don't think he knows about second breakfast, Pip.",
... "What about elevensies?",
... ]
>>> encoded_input = tokenizer(batch_sentences, padding=True, truncation=True, return_tensors="pt")
>>> print(encoded_input)
{'input_ids': tensor([[101, 1252, 1184, 1164, 1248, 6462, 136, 102, 0, 0, 0, 0, 0, 0, 0],
[101, 1790, 112, 189, 1341, 1119, 3520, 1164, 1248, 6462, 117, 21902, 1643, 119, 102],
[101, 1327, 1164, 5450, 23434, 136, 102, 0, 0, 0, 0, 0, 0, 0, 0]]),
'token_type_ids': tensor([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]),
'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0]])}
```
</pt>
<tf>
```py
>>> batch_sentences = [
... "But what about second breakfast?",
... "Don't think he knows about second breakfast, Pip.",
... "What about elevensies?",
... ]
>>> encoded_input = tokenizer(batch_sentences, padding=True, truncation=True, return_tensors="tf")
>>> print(encoded_input)
{'input_ids': <tf.Tensor: shape=(2, 9), dtype=int32, numpy=
array([[101, 1252, 1184, 1164, 1248, 6462, 136, 102, 0, 0, 0, 0, 0, 0, 0],
[101, 1790, 112, 189, 1341, 1119, 3520, 1164, 1248, 6462, 117, 21902, 1643, 119, 102],
[101, 1327, 1164, 5450, 23434, 136, 102, 0, 0, 0, 0, 0, 0, 0, 0]],
dtype=int32)>,
'token_type_ids': <tf.Tensor: shape=(2, 9), dtype=int32, numpy=
array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]], dtype=int32)>,
'attention_mask': <tf.Tensor: shape=(2, 9), dtype=int32, numpy=
array([[1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0]], dtype=int32)>}
```
</tf>
</frameworkcontent>
<Tip>
Different pipelines support tokenizer arguments in their `__call__()` differently. `text-2-text-generation` pipelines support (i.e. pass on)
only `truncation`. `text-generation` pipelines support `max_length`, `truncation`, `padding` and `add_special_tokens`.
In `fill-mask` pipelines, tokenizer arguments can be passed in the `tokenizer_kwargs` argument (dictionary).
</Tip>
## Audio
For audio tasks, you'll need a [feature extractor](main_classes/feature_extractor) to prepare your dataset for the model. The feature extractor is designed to extract features from raw audio data, and convert them into tensors.
Load the [MInDS-14](https://huggingface.co/datasets/PolyAI/minds14) dataset (see the ð€ [Datasets tutorial](https://huggingface.co/docs/datasets/load_hub) for more details on how to load a dataset) to see how you can use a feature extractor with audio datasets:
```py
>>> from datasets import load_dataset, Audio
>>> dataset = load_dataset("PolyAI/minds14", name="en-US", split="train")
```
Access the first element of the `audio` column to take a look at the input. Calling the `audio` column automatically loads and resamples the audio file:
```py
>>> dataset[0]["audio"]
{'array': array([ 0. , 0.00024414, -0.00024414, ..., -0.00024414,
0. , 0. ], dtype=float32),
'path': '/root/.cache/huggingface/datasets/downloads/extracted/f14948e0e84be638dd7943ac36518a4cf3324e8b7aa331c5ab11541518e9368c/en-US~JOINT_ACCOUNT/602ba55abb1e6d0fbce92065.wav',
'sampling_rate': 8000}
```
This returns three items:
* `array` is the speech signal loaded - and potentially resampled - as a 1D array.
* `path` points to the location of the audio file.
* `sampling_rate` refers to how many data points in the speech signal are measured per second.
For this tutorial, you'll use the [Wav2Vec2](https://huggingface.co/facebook/wav2vec2-base) model. Take a look at the model card, and you'll learn Wav2Vec2 is pretrained on 16kHz sampled speech audio. It is important your audio data's sampling rate matches the sampling rate of the dataset used to pretrain the model. If your data's sampling rate isn't the same, then you need to resample your data.
1. Use ð€ Datasets' [`~datasets.Dataset.cast_column`] method to upsample the sampling rate to 16kHz:
```py
>>> dataset = dataset.cast_column("audio", Audio(sampling_rate=16_000))
```
2. Call the `audio` column again to resample the audio file:
```py
>>> dataset[0]["audio"]
{'array': array([ 2.3443763e-05, 2.1729663e-04, 2.2145823e-04, ...,
3.8356509e-05, -7.3497440e-06, -2.1754686e-05], dtype=float32),
'path': '/root/.cache/huggingface/datasets/downloads/extracted/f14948e0e84be638dd7943ac36518a4cf3324e8b7aa331c5ab11541518e9368c/en-US~JOINT_ACCOUNT/602ba55abb1e6d0fbce92065.wav',
'sampling_rate': 16000}
```
Next, load a feature extractor to normalize and pad the input. When padding textual data, a `0` is added for shorter sequences. The same idea applies to audio data. The feature extractor adds a `0` - interpreted as silence - to `array`.
Load the feature extractor with [`AutoFeatureExtractor.from_pretrained`]:
```py
>>> from transformers import AutoFeatureExtractor
>>> feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/wav2vec2-base")
```
Pass the audio `array` to the feature extractor. We also recommend adding the `sampling_rate` argument in the feature extractor in order to better debug any silent errors that may occur.
```py
>>> audio_input = [dataset[0]["audio"]["array"]]
>>> feature_extractor(audio_input, sampling_rate=16000)
{'input_values': [array([ 3.8106556e-04, 2.7506407e-03, 2.8015103e-03, ...,
5.6335266e-04, 4.6588284e-06, -1.7142107e-04], dtype=float32)]}
```
Just like the tokenizer, you can apply padding or truncation to handle variable sequences in a batch. Take a look at the sequence length of these two audio samples:
```py
>>> dataset[0]["audio"]["array"].shape
(173398,)
>>> dataset[1]["audio"]["array"].shape
(106496,)
```
Create a function to preprocess the dataset so the audio samples are the same lengths. Specify a maximum sample length, and the feature extractor will either pad or truncate the sequences to match it:
```py
>>> def preprocess_function(examples):
... audio_arrays = [x["array"] for x in examples["audio"]]
... inputs = feature_extractor(
... audio_arrays,
... sampling_rate=16000,
... padding=True,
... max_length=100000,
... truncation=True,
... )
... return inputs
```
Apply the `preprocess_function` to the first few examples in the dataset:
```py
>>> processed_dataset = preprocess_function(dataset[:5])
```
The sample lengths are now the same and match the specified maximum length. You can pass your processed dataset to the model now!
```py
>>> processed_dataset["input_values"][0].shape
(100000,)
>>> processed_dataset["input_values"][1].shape
(100000,)
```
## Computer vision
For computer vision tasks, you'll need an [image processor](main_classes/image_processor) to prepare your dataset for the model.
Image preprocessing consists of several steps that convert images into the input expected by the model. These steps
include but are not limited to resizing, normalizing, color channel correction, and converting images to tensors.
<Tip>
Image preprocessing often follows some form of image augmentation. Both image preprocessing and image augmentation
transform image data, but they serve different purposes:
* Image augmentation alters images in a way that can help prevent overfitting and increase the robustness of the model. You can get creative in how you augment your data - adjust brightness and colors, crop, rotate, resize, zoom, etc. However, be mindful not to change the meaning of the images with your augmentations.
* Image preprocessing guarantees that the images match the modelâs expected input format. When fine-tuning a computer vision model, images must be preprocessed exactly as when the model was initially trained.
You can use any library you like for image augmentation. For image preprocessing, use the `ImageProcessor` associated with the model.
</Tip>
Load the [food101](https://huggingface.co/datasets/food101) dataset (see the ð€ [Datasets tutorial](https://huggingface.co/docs/datasets/load_hub) for more details on how to load a dataset) to see how you can use an image processor with computer vision datasets:
<Tip>
Use ð€ Datasets `split` parameter to only load a small sample from the training split since the dataset is quite large!
</Tip>
```py
>>> from datasets import load_dataset
>>> dataset = load_dataset("food101", split="train[:100]")
```
Next, take a look at the image with ð€ Datasets [`Image`](https://huggingface.co/docs/datasets/package_reference/main_classes?highlight=image#datasets.Image) feature:
```py
>>> dataset[0]["image"]
```
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/vision-preprocess-tutorial.png"/>
</div>
Load the image processor with [`AutoImageProcessor.from_pretrained`]:
```py
>>> from transformers import AutoImageProcessor
>>> image_processor = AutoImageProcessor.from_pretrained("google/vit-base-patch16-224")
```
First, let's add some image augmentation. You can use any library you prefer, but in this tutorial, we'll use torchvision's [`transforms`](https://pytorch.org/vision/stable/transforms.html) module. If you're interested in using another data augmentation library, learn how in the [Albumentations](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/image_classification_albumentations.ipynb) or [Kornia notebooks](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/image_classification_kornia.ipynb).
1. Here we use [`Compose`](https://pytorch.org/vision/master/generated/torchvision.transforms.Compose.html) to chain together a couple of
transforms - [`RandomResizedCrop`](https://pytorch.org/vision/main/generated/torchvision.transforms.RandomResizedCrop.html) and [`ColorJitter`](https://pytorch.org/vision/main/generated/torchvision.transforms.ColorJitter.html).
Note that for resizing, we can get the image size requirements from the `image_processor`. For some models, an exact height and
width are expected, for others only the `shortest_edge` is defined.
```py
>>> from torchvision.transforms import RandomResizedCrop, ColorJitter, Compose
>>> size = (
... image_processor.size["shortest_edge"]
... if "shortest_edge" in image_processor.size
... else (image_processor.size["height"], image_processor.size["width"])
... )
>>> _transforms = Compose([RandomResizedCrop(size), ColorJitter(brightness=0.5, hue=0.5)])
```
2. The model accepts [`pixel_values`](model_doc/vision-encoder-decoder#transformers.VisionEncoderDecoderModel.forward.pixel_values)
as its input. `ImageProcessor` can take care of normalizing the images, and generating appropriate tensors.
Create a function that combines image augmentation and image preprocessing for a batch of images and generates `pixel_values`:
```py
>>> def transforms(examples):
... images = [_transforms(img.convert("RGB")) for img in examples["image"]]
... examples["pixel_values"] = image_processor(images, do_resize=False, return_tensors="pt")["pixel_values"]
... return examples
```
<Tip>
In the example above we set `do_resize=False` because we have already resized the images in the image augmentation transformation,
and leveraged the `size` attribute from the appropriate `image_processor`. If you do not resize images during image augmentation,
leave this parameter out. By default, `ImageProcessor` will handle the resizing.
If you wish to normalize images as a part of the augmentation transformation, use the `image_processor.image_mean`,
and `image_processor.image_std` values.
</Tip>
3. Then use ð€ Datasets[`~datasets.Dataset.set_transform`] to apply the transforms on the fly:
```py
>>> dataset.set_transform(transforms)
```
4. Now when you access the image, you'll notice the image processor has added `pixel_values`. You can pass your processed dataset to the model now!
```py
>>> dataset[0].keys()
```
Here is what the image looks like after the transforms are applied. The image has been randomly cropped and it's color properties are different.
```py
>>> import numpy as np
>>> import matplotlib.pyplot as plt
>>> img = dataset[0]["pixel_values"]
>>> plt.imshow(img.permute(1, 2, 0))
```
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/preprocessed_image.png"/>
</div>
<Tip>
For tasks like object detection, semantic segmentation, instance segmentation, and panoptic segmentation, `ImageProcessor`
offers post processing methods. These methods convert model's raw outputs into meaningful predictions such as bounding boxes,
or segmentation maps.
</Tip>
### Pad
In some cases, for instance, when fine-tuning [DETR](./model_doc/detr), the model applies scale augmentation at training
time. This may cause images to be different sizes in a batch. You can use [`DetrImageProcessor.pad`]
from [`DetrImageProcessor`] and define a custom `collate_fn` to batch images together.
```py
>>> def collate_fn(batch):
... pixel_values = [item["pixel_values"] for item in batch]
... encoding = image_processor.pad(pixel_values, return_tensors="pt")
... labels = [item["labels"] for item in batch]
... batch = {}
... batch["pixel_values"] = encoding["pixel_values"]
... batch["pixel_mask"] = encoding["pixel_mask"]
... batch["labels"] = labels
... return batch
```
## Multimodal
For tasks involving multimodal inputs, you'll need a [processor](main_classes/processors) to prepare your dataset for the model. A processor couples together two processing objects such as as tokenizer and feature extractor.
Load the [LJ Speech](https://huggingface.co/datasets/lj_speech) dataset (see the ð€ [Datasets tutorial](https://huggingface.co/docs/datasets/load_hub) for more details on how to load a dataset) to see how you can use a processor for automatic speech recognition (ASR):
```py
>>> from datasets import load_dataset
>>> lj_speech = load_dataset("lj_speech", split="train")
```
For ASR, you're mainly focused on `audio` and `text` so you can remove the other columns:
```py
>>> lj_speech = lj_speech.map(remove_columns=["file", "id", "normalized_text"])
```
Now take a look at the `audio` and `text` columns:
```py
>>> lj_speech[0]["audio"]
{'array': array([-7.3242188e-04, -7.6293945e-04, -6.4086914e-04, ...,
7.3242188e-04, 2.1362305e-04, 6.1035156e-05], dtype=float32),
'path': '/root/.cache/huggingface/datasets/downloads/extracted/917ece08c95cf0c4115e45294e3cd0dee724a1165b7fc11798369308a465bd26/LJSpeech-1.1/wavs/LJ001-0001.wav',
'sampling_rate': 22050}
>>> lj_speech[0]["text"]
'Printing, in the only sense with which we are at present concerned, differs from most if not from all the arts and crafts represented in the Exhibition'
```
Remember you should always [resample](preprocessing#audio) your audio dataset's sampling rate to match the sampling rate of the dataset used to pretrain a model!
```py
>>> lj_speech = lj_speech.cast_column("audio", Audio(sampling_rate=16_000))
```
Load a processor with [`AutoProcessor.from_pretrained`]:
```py
>>> from transformers import AutoProcessor
>>> processor = AutoProcessor.from_pretrained("facebook/wav2vec2-base-960h")
```
1. Create a function to process the audio data contained in `array` to `input_values`, and tokenize `text` to `labels`. These are the inputs to the model:
```py
>>> def prepare_dataset(example):
... audio = example["audio"]
... example.update(processor(audio=audio["array"], text=example["text"], sampling_rate=16000))
... return example
```
2. Apply the `prepare_dataset` function to a sample:
```py
>>> prepare_dataset(lj_speech[0])
```
The processor has now added `input_values` and `labels`, and the sampling rate has also been correctly downsampled to 16kHz. You can pass your processed dataset to the model now!
| transformers/docs/source/en/preprocessing.md/0 | {
"file_path": "transformers/docs/source/en/preprocessing.md",
"repo_id": "transformers",
"token_count": 8685
} | 244 |
<!--Copyright 2023 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# How ð€ Transformers solve tasks
In [What ð€ Transformers can do](task_summary), you learned about natural language processing (NLP), speech and audio, computer vision tasks, and some important applications of them. This page will look closely at how models solve these tasks and explain what's happening under the hood. There are many ways to solve a given task, some models may implement certain techniques or even approach the task from a new angle, but for Transformer models, the general idea is the same. Owing to its flexible architecture, most models are a variant of an encoder, decoder, or encoder-decoder structure. In addition to Transformer models, our library also has several convolutional neural networks (CNNs), which are still used today for computer vision tasks. We'll also explain how a modern CNN works.
To explain how tasks are solved, we'll walk through what goes on inside the model to output useful predictions.
- [Wav2Vec2](model_doc/wav2vec2) for audio classification and automatic speech recognition (ASR)
- [Vision Transformer (ViT)](model_doc/vit) and [ConvNeXT](model_doc/convnext) for image classification
- [DETR](model_doc/detr) for object detection
- [Mask2Former](model_doc/mask2former) for image segmentation
- [GLPN](model_doc/glpn) for depth estimation
- [BERT](model_doc/bert) for NLP tasks like text classification, token classification and question answering that use an encoder
- [GPT2](model_doc/gpt2) for NLP tasks like text generation that use a decoder
- [BART](model_doc/bart) for NLP tasks like summarization and translation that use an encoder-decoder
<Tip>
Before you go further, it is good to have some basic knowledge of the original Transformer architecture. Knowing how encoders, decoders, and attention work will aid you in understanding how different Transformer models work. If you're just getting started or need a refresher, check out our [course](https://huggingface.co/course/chapter1/4?fw=pt) for more information!
</Tip>
## Speech and audio
[Wav2Vec2](model_doc/wav2vec2) is a self-supervised model pretrained on unlabeled speech data and finetuned on labeled data for audio classification and automatic speech recognition.
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/wav2vec2_architecture.png"/>
</div>
This model has four main components:
1. A *feature encoder* takes the raw audio waveform, normalizes it to zero mean and unit variance, and converts it into a sequence of feature vectors that are each 20ms long.
2. Waveforms are continuous by nature, so they can't be divided into separate units like a sequence of text can be split into words. That's why the feature vectors are passed to a *quantization module*, which aims to learn discrete speech units. The speech unit is chosen from a collection of codewords, known as a *codebook* (you can think of this as the vocabulary). From the codebook, the vector or speech unit, that best represents the continuous audio input is chosen and forwarded through the model.
3. About half of the feature vectors are randomly masked, and the masked feature vector is fed to a *context network*, which is a Transformer encoder that also adds relative positional embeddings.
4. The pretraining objective of the context network is a *contrastive task*. The model has to predict the true quantized speech representation of the masked prediction from a set of false ones, encouraging the model to find the most similar context vector and quantized speech unit (the target label).
Now that wav2vec2 is pretrained, you can finetune it on your data for audio classification or automatic speech recognition!
### Audio classification
To use the pretrained model for audio classification, add a sequence classification head on top of the base Wav2Vec2 model. The classification head is a linear layer that accepts the encoder's hidden states. The hidden states represent the learned features from each audio frame which can have varying lengths. To create one vector of fixed-length, the hidden states are pooled first and then transformed into logits over the class labels. The cross-entropy loss is calculated between the logits and target to find the most likely class.
Ready to try your hand at audio classification? Check out our complete [audio classification guide](tasks/audio_classification) to learn how to finetune Wav2Vec2 and use it for inference!
### Automatic speech recognition
To use the pretrained model for automatic speech recognition, add a language modeling head on top of the base Wav2Vec2 model for [connectionist temporal classification (CTC)](glossary#connectionist-temporal-classification-ctc). The language modeling head is a linear layer that accepts the encoder's hidden states and transforms them into logits. Each logit represents a token class (the number of tokens comes from the task vocabulary). The CTC loss is calculated between the logits and targets to find the most likely sequence of tokens, which are then decoded into a transcription.
Ready to try your hand at automatic speech recognition? Check out our complete [automatic speech recognition guide](tasks/asr) to learn how to finetune Wav2Vec2 and use it for inference!
## Computer vision
There are two ways to approach computer vision tasks:
1. Split an image into a sequence of patches and process them in parallel with a Transformer.
2. Use a modern CNN, like [ConvNeXT](model_doc/convnext), which relies on convolutional layers but adopts modern network designs.
<Tip>
A third approach mixes Transformers with convolutions (for example, [Convolutional Vision Transformer](model_doc/cvt) or [LeViT](model_doc/levit)). We won't discuss those because they just combine the two approaches we examine here.
</Tip>
ViT and ConvNeXT are commonly used for image classification, but for other vision tasks like object detection, segmentation, and depth estimation, we'll look at DETR, Mask2Former and GLPN, respectively; these models are better suited for those tasks.
### Image classification
ViT and ConvNeXT can both be used for image classification; the main difference is that ViT uses an attention mechanism while ConvNeXT uses convolutions.
#### Transformer
[ViT](model_doc/vit) replaces convolutions entirely with a pure Transformer architecture. If you're familiar with the original Transformer, then you're already most of the way toward understanding ViT.
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/vit_architecture.jpg"/>
</div>
The main change ViT introduced was in how images are fed to a Transformer:
1. An image is split into square non-overlapping patches, each of which gets turned into a vector or *patch embedding*. The patch embeddings are generated from a convolutional 2D layer which creates the proper input dimensions (which for a base Transformer is 768 values for each patch embedding). If you had a 224x224 pixel image, you could split it into 196 16x16 image patches. Just like how text is tokenized into words, an image is "tokenized" into a sequence of patches.
2. A *learnable embedding* - a special `[CLS]` token - is added to the beginning of the patch embeddings just like BERT. The final hidden state of the `[CLS]` token is used as the input to the attached classification head; other outputs are ignored. This token helps the model learn how to encode a representation of the image.
3. The last thing to add to the patch and learnable embeddings are the *position embeddings* because the model doesn't know how the image patches are ordered. The position embeddings are also learnable and have the same size as the patch embeddings. Finally, all of the embeddings are passed to the Transformer encoder.
4. The output, specifically only the output with the `[CLS]` token, is passed to a multilayer perceptron head (MLP). ViT's pretraining objective is simply classification. Like other classification heads, the MLP head converts the output into logits over the class labels and calculates the cross-entropy loss to find the most likely class.
Ready to try your hand at image classification? Check out our complete [image classification guide](tasks/image_classification) to learn how to finetune ViT and use it for inference!
#### CNN
<Tip>
This section briefly explains convolutions, but it'd be helpful to have a prior understanding of how they change an image's shape and size. If you're unfamiliar with convolutions, check out the [Convolution Neural Networks chapter](https://github.com/fastai/fastbook/blob/master/13_convolutions.ipynb) from the fastai book!
</Tip>
[ConvNeXT](model_doc/convnext) is a CNN architecture that adopts new and modern network designs to improve performance. However, convolutions are still at the core of the model. From a high-level perspective, a [convolution](glossary#convolution) is an operation where a smaller matrix (*kernel*) is multiplied by a small window of the image pixels. It computes some features from it, such as a particular texture or curvature of a line. Then it slides over to the next window of pixels; the distance the convolution travels is known as the *stride*.
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/convolution.gif"/>
</div>
<small>A basic convolution without padding or stride, taken from <a href="https://arxiv.org/abs/1603.07285">A guide to convolution arithmetic for deep learning.</a></small>
You can feed this output to another convolutional layer, and with each successive layer, the network learns more complex and abstract things like hotdogs or rockets. Between convolutional layers, it is common to add a pooling layer to reduce dimensionality and make the model more robust to variations of a feature's position.
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/convnext_architecture.png"/>
</div>
ConvNeXT modernizes a CNN in five ways:
1. Change the number of blocks in each stage and "patchify" an image with a larger stride and corresponding kernel size. The non-overlapping sliding window makes this patchifying strategy similar to how ViT splits an image into patches.
2. A *bottleneck* layer shrinks the number of channels and then restores it because it is faster to do a 1x1 convolution, and you can increase the depth. An inverted bottleneck does the opposite by expanding the number of channels and shrinking them, which is more memory efficient.
3. Replace the typical 3x3 convolutional layer in the bottleneck layer with *depthwise convolution*, which applies a convolution to each input channel separately and then stacks them back together at the end. This widens the network width for improved performance.
4. ViT has a global receptive field which means it can see more of an image at once thanks to its attention mechanism. ConvNeXT attempts to replicate this effect by increasing the kernel size to 7x7.
5. ConvNeXT also makes several layer design changes that imitate Transformer models. There are fewer activation and normalization layers, the activation function is switched to GELU instead of ReLU, and it uses LayerNorm instead of BatchNorm.
The output from the convolution blocks is passed to a classification head which converts the outputs into logits and calculates the cross-entropy loss to find the most likely label.
### Object detection
[DETR](model_doc/detr), *DEtection TRansformer*, is an end-to-end object detection model that combines a CNN with a Transformer encoder-decoder.
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/detr_architecture.png"/>
</div>
1. A pretrained CNN *backbone* takes an image, represented by its pixel values, and creates a low-resolution feature map of it. A 1x1 convolution is applied to the feature map to reduce dimensionality and it creates a new feature map with a high-level image representation. Since the Transformer is a sequential model, the feature map is flattened into a sequence of feature vectors that are combined with positional embeddings.
2. The feature vectors are passed to the encoder, which learns the image representations using its attention layers. Next, the encoder hidden states are combined with *object queries* in the decoder. Object queries are learned embeddings that focus on the different regions of an image, and they're updated as they progress through each attention layer. The decoder hidden states are passed to a feedforward network that predicts the bounding box coordinates and class label for each object query, or `no object` if there isn't one.
DETR decodes each object query in parallel to output *N* final predictions, where *N* is the number of queries. Unlike a typical autoregressive model that predicts one element at a time, object detection is a set prediction task (`bounding box`, `class label`) that makes *N* predictions in a single pass.
3. DETR uses a *bipartite matching loss* during training to compare a fixed number of predictions with a fixed set of ground truth labels. If there are fewer ground truth labels in the set of *N* labels, then they're padded with a `no object` class. This loss function encourages DETR to find a one-to-one assignment between the predictions and ground truth labels. If either the bounding boxes or class labels aren't correct, a loss is incurred. Likewise, if DETR predicts an object that doesn't exist, it is penalized. This encourages DETR to find other objects in an image instead of focusing on one really prominent object.
An object detection head is added on top of DETR to find the class label and the coordinates of the bounding box. There are two components to the object detection head: a linear layer to transform the decoder hidden states into logits over the class labels, and a MLP to predict the bounding box.
Ready to try your hand at object detection? Check out our complete [object detection guide](tasks/object_detection) to learn how to finetune DETR and use it for inference!
### Image segmentation
[Mask2Former](model_doc/mask2former) is a universal architecture for solving all types of image segmentation tasks. Traditional segmentation models are typically tailored towards a particular subtask of image segmentation, like instance, semantic or panoptic segmentation. Mask2Former frames each of those tasks as a *mask classification* problem. Mask classification groups pixels into *N* segments, and predicts *N* masks and their corresponding class label for a given image. We'll explain how Mask2Former works in this section, and then you can try finetuning SegFormer at the end.
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/mask2former_architecture.png"/>
</div>
There are three main components to Mask2Former:
1. A [Swin](model_doc/swin) backbone accepts an image and creates a low-resolution image feature map from 3 consecutive 3x3 convolutions.
2. The feature map is passed to a *pixel decoder* which gradually upsamples the low-resolution features into high-resolution per-pixel embeddings. The pixel decoder actually generates multi-scale features (contains both low- and high-resolution features) with resolutions 1/32, 1/16, and 1/8th of the original image.
3. Each of these feature maps of differing scales is fed successively to one Transformer decoder layer at a time in order to capture small objects from the high-resolution features. The key to Mask2Former is the *masked attention* mechanism in the decoder. Unlike cross-attention which can attend to the entire image, masked attention only focuses on a certain area of the image. This is faster and leads to better performance because the local features of an image are enough for the model to learn from.
4. Like [DETR](tasks_explained#object-detection), Mask2Former also uses learned object queries and combines them with the image features from the pixel decoder to make a set prediction (`class label`, `mask prediction`). The decoder hidden states are passed into a linear layer and transformed into logits over the class labels. The cross-entropy loss is calculated between the logits and class label to find the most likely one.
The mask predictions are generated by combining the pixel-embeddings with the final decoder hidden states. The sigmoid cross-entropy and dice loss is calculated between the logits and the ground truth mask to find the most likely mask.
Ready to try your hand at object detection? Check out our complete [image segmentation guide](tasks/semantic_segmentation) to learn how to finetune SegFormer and use it for inference!
### Depth estimation
[GLPN](model_doc/glpn), *Global-Local Path Network*, is a Transformer for depth estimation that combines a [SegFormer](model_doc/segformer) encoder with a lightweight decoder.
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/glpn_architecture.jpg"/>
</div>
1. Like ViT, an image is split into a sequence of patches, except these image patches are smaller. This is better for dense prediction tasks like segmentation or depth estimation. The image patches are transformed into patch embeddings (see the [image classification](#image-classification) section for more details about how patch embeddings are created), which are fed to the encoder.
2. The encoder accepts the patch embeddings, and passes them through several encoder blocks. Each block consists of attention and Mix-FFN layers. The purpose of the latter is to provide positional information. At the end of each encoder block is a *patch merging* layer for creating hierarchical representations. The features of each group of neighboring patches are concatenated, and a linear layer is applied to the concatenated features to reduce the number of patches to a resolution of 1/4. This becomes the input to the next encoder block, where this whole process is repeated until you have image features with resolutions of 1/8, 1/16, and 1/32.
3. A lightweight decoder takes the last feature map (1/32 scale) from the encoder and upsamples it to 1/16 scale. From here, the feature is passed into a *Selective Feature Fusion (SFF)* module, which selects and combines local and global features from an attention map for each feature and then upsamples it to 1/8th. This process is repeated until the decoded features are the same size as the original image. The output is passed through two convolution layers and then a sigmoid activation is applied to predict the depth of each pixel.
## Natural language processing
The Transformer was initially designed for machine translation, and since then, it has practically become the default architecture for solving all NLP tasks. Some tasks lend themselves to the Transformer's encoder structure, while others are better suited for the decoder. Still, other tasks make use of both the Transformer's encoder-decoder structure.
### Text classification
[BERT](model_doc/bert) is an encoder-only model and is the first model to effectively implement deep bidirectionality to learn richer representations of the text by attending to words on both sides.
1. BERT uses [WordPiece](tokenizer_summary#wordpiece) tokenization to generate a token embedding of the text. To tell the difference between a single sentence and a pair of sentences, a special `[SEP]` token is added to differentiate them. A special `[CLS]` token is added to the beginning of every sequence of text. The final output with the `[CLS]` token is used as the input to the classification head for classification tasks. BERT also adds a segment embedding to denote whether a token belongs to the first or second sentence in a pair of sentences.
2. BERT is pretrained with two objectives: masked language modeling and next-sentence prediction. In masked language modeling, some percentage of the input tokens are randomly masked, and the model needs to predict these. This solves the issue of bidirectionality, where the model could cheat and see all the words and "predict" the next word. The final hidden states of the predicted mask tokens are passed to a feedforward network with a softmax over the vocabulary to predict the masked word.
The second pretraining object is next-sentence prediction. The model must predict whether sentence B follows sentence A. Half of the time sentence B is the next sentence, and the other half of the time, sentence B is a random sentence. The prediction, whether it is the next sentence or not, is passed to a feedforward network with a softmax over the two classes (`IsNext` and `NotNext`).
3. The input embeddings are passed through multiple encoder layers to output some final hidden states.
To use the pretrained model for text classification, add a sequence classification head on top of the base BERT model. The sequence classification head is a linear layer that accepts the final hidden states and performs a linear transformation to convert them into logits. The cross-entropy loss is calculated between the logits and target to find the most likely label.
Ready to try your hand at text classification? Check out our complete [text classification guide](tasks/sequence_classification) to learn how to finetune DistilBERT and use it for inference!
### Token classification
To use BERT for token classification tasks like named entity recognition (NER), add a token classification head on top of the base BERT model. The token classification head is a linear layer that accepts the final hidden states and performs a linear transformation to convert them into logits. The cross-entropy loss is calculated between the logits and each token to find the most likely label.
Ready to try your hand at token classification? Check out our complete [token classification guide](tasks/token_classification) to learn how to finetune DistilBERT and use it for inference!
### Question answering
To use BERT for question answering, add a span classification head on top of the base BERT model. This linear layer accepts the final hidden states and performs a linear transformation to compute the `span` start and end logits corresponding to the answer. The cross-entropy loss is calculated between the logits and the label position to find the most likely span of text corresponding to the answer.
Ready to try your hand at question answering? Check out our complete [question answering guide](tasks/question_answering) to learn how to finetune DistilBERT and use it for inference!
<Tip>
ð¡ Notice how easy it is to use BERT for different tasks once it's been pretrained. You only need to add a specific head to the pretrained model to manipulate the hidden states into your desired output!
</Tip>
### Text generation
[GPT-2](model_doc/gpt2) is a decoder-only model pretrained on a large amount of text. It can generate convincing (though not always true!) text given a prompt and complete other NLP tasks like question answering despite not being explicitly trained to.
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/gpt2_architecture.png"/>
</div>
1. GPT-2 uses [byte pair encoding (BPE)](tokenizer_summary#bytepair-encoding-bpe) to tokenize words and generate a token embedding. Positional encodings are added to the token embeddings to indicate the position of each token in the sequence. The input embeddings are passed through multiple decoder blocks to output some final hidden state. Within each decoder block, GPT-2 uses a *masked self-attention* layer which means GPT-2 can't attend to future tokens. It is only allowed to attend to tokens on the left. This is different from BERT's [`mask`] token because, in masked self-attention, an attention mask is used to set the score to `0` for future tokens.
2. The output from the decoder is passed to a language modeling head, which performs a linear transformation to convert the hidden states into logits. The label is the next token in the sequence, which are created by shifting the logits to the right by one. The cross-entropy loss is calculated between the shifted logits and the labels to output the next most likely token.
GPT-2's pretraining objective is based entirely on [causal language modeling](glossary#causal-language-modeling), predicting the next word in a sequence. This makes GPT-2 especially good at tasks that involve generating text.
Ready to try your hand at text generation? Check out our complete [causal language modeling guide](tasks/language_modeling#causal-language-modeling) to learn how to finetune DistilGPT-2 and use it for inference!
<Tip>
For more information about text generation, check out the [text generation strategies](generation_strategies) guide!
</Tip>
### Summarization
Encoder-decoder models like [BART](model_doc/bart) and [T5](model_doc/t5) are designed for the sequence-to-sequence pattern of a summarization task. We'll explain how BART works in this section, and then you can try finetuning T5 at the end.
<div class="flex justify-center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bart_architecture.png"/>
</div>
1. BART's encoder architecture is very similar to BERT and accepts a token and positional embedding of the text. BART is pretrained by corrupting the input and then reconstructing it with the decoder. Unlike other encoders with specific corruption strategies, BART can apply any type of corruption. The *text infilling* corruption strategy works the best though. In text infilling, a number of text spans are replaced with a **single** [`mask`] token. This is important because the model has to predict the masked tokens, and it teaches the model to predict the number of missing tokens. The input embeddings and masked spans are passed through the encoder to output some final hidden states, but unlike BERT, BART doesn't add a final feedforward network at the end to predict a word.
2. The encoder's output is passed to the decoder, which must predict the masked tokens and any uncorrupted tokens from the encoder's output. This gives additional context to help the decoder restore the original text. The output from the decoder is passed to a language modeling head, which performs a linear transformation to convert the hidden states into logits. The cross-entropy loss is calculated between the logits and the label, which is just the token shifted to the right.
Ready to try your hand at summarization? Check out our complete [summarization guide](tasks/summarization) to learn how to finetune T5 and use it for inference!
<Tip>
For more information about text generation, check out the [text generation strategies](generation_strategies) guide!
</Tip>
### Translation
Translation is another example of a sequence-to-sequence task, which means you can use an encoder-decoder model like [BART](model_doc/bart) or [T5](model_doc/t5) to do it. We'll explain how BART works in this section, and then you can try finetuning T5 at the end.
BART adapts to translation by adding a separate randomly initialized encoder to map a source language to an input that can be decoded into the target language. This new encoder's embeddings are passed to the pretrained encoder instead of the original word embeddings. The source encoder is trained by updating the source encoder, positional embeddings, and input embeddings with the cross-entropy loss from the model output. The model parameters are frozen in this first step, and all the model parameters are trained together in the second step.
BART has since been followed up by a multilingual version, mBART, intended for translation and pretrained on many different languages.
Ready to try your hand at translation? Check out our complete [translation guide](tasks/summarization) to learn how to finetune T5 and use it for inference!
<Tip>
For more information about text generation, check out the [text generation strategies](generation_strategies) guide!
</Tip> | transformers/docs/source/en/tasks_explained.md/0 | {
"file_path": "transformers/docs/source/en/tasks_explained.md",
"repo_id": "transformers",
"token_count": 6963
} | 245 |
<!---
Copyright 2020 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# Verificaciones en un Pull Request
Cuando abres un _pull request_ en ð€ Transformers, se ejecutarán una serie de verificaciones para asegurarte de que el _patch_ que estás agregando no rompa nada existente. Estas verificaciones son de cuatro tipos:
- pruebas regulares
- creación de la documentación
- estilo del código y documentación
- consistencia del repositorio
En este documento, intentaremos explicar cuáles son esas diferentes verificaciones y el motivo detrás de ellas, asà como también cómo depurarlas localmente si una falla en tu PR.
Recuerda que todas las verificaciones requieren que tengas una instalación de desarrollo:
```bash
pip install transformers[dev]
```
o una instalación editable:
```bash
pip install -e .[dev]
```
del repositorio de Transformers.
## Pruebas
Todos los procesos que comienzan con `ci/circleci: run_tests_` ejecutan partes del conjunto de pruebas de Transformers. Cada uno de esos procesos se enfoca en una parte de la biblioteca en un entorno determinado: por ejemplo, `ci/circleci: run_tests_pipelines_tf` ejecuta la prueba de _pipelines_ en un entorno donde solo está instalado TensorFlow.
Ten en cuenta que para evitar ejecutar pruebas cuando no hay un cambio real en los módulos que estás probando, solo se ejecuta una parte del conjunto de pruebas: se ejecuta una tarea auxiliar para determinar las diferencias en la biblioteca antes y después del PR (lo que GitHub te muestra en la pestaña "Files changes") y selecciona las pruebas afectadas por esa diferencia. Este auxiliar se puede ejecutar localmente usando:
```bash
python utils/tests_fetcher.py
```
desde el directorio raiz del repositorio de Transformers. Se ejecutará lo siguiente:
1. Verificación para cada archivo en el _diff_ si los cambios están en el código, solo en comentarios o _docstrings_. Solo los archivos con cambios reales de código se conservan.
2. Creación de un mapa interno que proporciona para cada archivo del código fuente de la biblioteca todos los archivos a los que impacta recursivamente. Se dice que el módulo A impacta al módulo B si el módulo B importa el módulo A. Para el impacto recursivo, necesitamos una cadena de módulos que va del módulo A al módulo B en la que cada módulo importa el anterior.
3. Aplicación de este mapa en los archivos recopilados en el paso 1, lo que nos da una lista de archivos modelo afectados por el PR.
4. Asignación de cada uno de esos archivos a sus archivos de prueba correspondientes y para obtener una la lista de pruebas a ejecutar.
Al ejecutar el _script_ localmente, debes obtener los resultados de los pasos 1, 3 y 4 impresos y asà saber qué pruebas se ejecutarán. El _script_ también creará un archivo llamado `test_list.txt` que contiene la lista de pruebas para ejecutar, y puede ejecutarlas localmente con el siguiente comando:
```bash
python -m pytest -n 8 --dist=loadfile -rA -s $(cat test_list.txt)
```
En caso de que se te escape algo, el conjunto completo de pruebas también se ejecuta a diario.
## Creación de la documentación
El proceso `build_pr_documentation` compila y genera una vista previa de la documentación para asegurarse de que todo se vea bien una vez que se fusione tu PR. Un bot agregará un enlace para obtener una vista previa de la documentación en tu PR. Cualquier cambio que realices en el PR se actualiza automáticamente en la vista previa. Si la documentación no se genera, haz clic en **Detalles** junto al proceso fallido para ver dónde salió mal. A menudo, el error es tan simple como que falta un archivo en `toctree`.
Si estás interesado en compilar u obtener una vista previa de la documentación localmente, echa un vistazo al [`README.md`](https://github.com/huggingface/transformers/tree/main/docs) en la carpeta `docs`.
## Estilo de código y documentación.
El formato de código se aplica a todos los archivos fuente, los ejemplos y las pruebas utilizando `black` e `ruff`. También tenemos una herramienta personalizada que se ocupa del formato de los _docstrings_ y archivos `rst` (`utils/style_doc.py`), asà como del orden de las importaciones _lazy_ realizadas en los archivos `__init__.py` de Transformers (`utils /custom_init_isort.py`). Todo esto se puede probar ejecutando
```bash
make style
```
CI verifica que se hayan aplicado dentro de la verificación `ci/circleci: check_code_quality`. También se ejecuta `ruff`, que hará una verificación básica a tu código y te hará saber si encuentra una variable no definida, o una que no se usa. Para ejecutar esa verificación localmente, usa
```bash
make quality
```
Esto puede llevar mucho tiempo, asà que para ejecutar lo mismo solo en los archivos que modificaste en la rama actual, ejecuta
```bash
make fixup
```
Este último comando también ejecutará todas las verificaciones adicionales para la consistencia del repositorio. Echemos un vistazo a estas pruebas.
## Consistencia del repositorio
Esta verificación reagrupa todas las pruebas para asegurarse de que tu PR deja el repositorio en buen estado, y se realiza mediante `ci/circleci: check_repository_consistency`. Puedes ejecutar localmente esta verificación ejecutando lo siguiente:
```bash
make repo-consistency
```
Esta instrucción verifica que:
- Todos los objetos agregados al _init_ están documentados (realizados por `utils/check_repo.py`)
- Todos los archivos `__init__.py` tienen el mismo contenido en sus dos secciones (realizado por `utils/check_inits.py`)
- Todo el código identificado como una copia de otro módulo es consistente con el original (realizado por `utils/check_copies.py`)
- Todas las clases de configuración tienen al menos _checkpoint_ válido mencionado en sus _docstrings_ (realizado por `utils/check_config_docstrings.py`)
- Las traducciones de los README y el Ãndice del documento tienen la misma lista de modelos que el README principal (realizado por `utils/check_copies.py`)
- Las tablas generadas automaticamente en la documentación están actualizadas (realizadas por `utils/check_table.py`)
- La biblioteca tiene todos los objetos disponibles incluso si no están instaladas todas las dependencias opcionales (realizadas por `utils/check_dummies.py`)
Si esta verificación falla, los primeros dos elementos requieren una reparación manual, los últimos cuatro pueden repararse automáticamente ejecutando el comando
```bash
make fix-copies
```
Las verificaciones adicionales se refieren a los PRs que agregan nuevos modelos, principalmente que:
- Todos los modelos agregados están en un Auto-mapping (realizado por `utils/check_repo.py`)
<!-- TODO Sylvain, add a check that makes sure the common tests are implemented.-->
- Todos los modelos se verifican correctamente (realizados por `utils/check_repo.py`)
<!-- TODO Sylvain, add the following
- All models are added to the main README, inside the main doc
- All checkpoints used actually exist on the Hub
-->
| transformers/docs/source/es/pr_checks.md/0 | {
"file_path": "transformers/docs/source/es/pr_checks.md",
"repo_id": "transformers",
"token_count": 2659
} | 246 |
<!--â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# Traduction en cours. | transformers/docs/source/fr/in_translation.md/0 | {
"file_path": "transformers/docs/source/fr/in_translation.md",
"repo_id": "transformers",
"token_count": 54
} | 247 |
<!--Copyright 2020 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# Condividere modelli personalizzati
La libreria ð€ Transformers Ú studiata per essere facilmente estendibile. Il codice di ogni modello Ú interamente
situato in una sottocartella del repository senza alcuna astrazione, perciò puoi facilmente copiare il file di un
modello e modificarlo in base ai tuoi bisogni.
Se stai scrivendo un nuovo modello, potrebbe essere più semplice iniziare da zero. In questo tutorial, ti mostreremo
come scrivere un modello personalizzato e la sua configurazione in modo che possa essere utilizzato allâinterno di
Transformers, e come condividerlo con la community (assieme al relativo codice) così che tutte le persone possano usarlo, anche
se non presente nella libreria ð€ Transformers.
Illustriamo tutto questo su un modello ResNet, avvolgendo la classe ResNet della
[libreria timm](https://github.com/rwightman/pytorch-image-models) in un [`PreTrainedModel`].
## Scrivere una configurazione personalizzata
Prima di iniziare a lavorare al modello, scriviamone la configurazione. La configurazione di un modello Ú un oggetto
che contiene tutte le informazioni necessarie per la build del modello. Come vedremo nella prossima sezione, il
modello può soltanto essere inizializzato tramite `config`, per cui dovremo rendere tale oggetto più completo possibile.
Nel nostro esempio, prenderemo un paio di argomenti della classe ResNet che potremmo voler modificare.
Configurazioni differenti ci daranno quindi i differenti possibili tipi di ResNet. Salveremo poi questi argomenti,
dopo averne controllato la validità .
```python
from transformers import PretrainedConfig
from typing import List
class ResnetConfig(PretrainedConfig):
model_type = "resnet"
def __init__(
self,
block_type="bottleneck",
layers: List[int] = [3, 4, 6, 3],
num_classes: int = 1000,
input_channels: int = 3,
cardinality: int = 1,
base_width: int = 64,
stem_width: int = 64,
stem_type: str = "",
avg_down: bool = False,
**kwargs,
):
if block_type not in ["basic", "bottleneck"]:
raise ValueError(f"`block_type` must be 'basic' or bottleneck', got {block_type}.")
if stem_type not in ["", "deep", "deep-tiered"]:
raise ValueError(f"`stem_type` must be '', 'deep' or 'deep-tiered', got {stem_type}.")
self.block_type = block_type
self.layers = layers
self.num_classes = num_classes
self.input_channels = input_channels
self.cardinality = cardinality
self.base_width = base_width
self.stem_width = stem_width
self.stem_type = stem_type
self.avg_down = avg_down
super().__init__(**kwargs)
```
Le tre cose più importanti da ricordare quando scrivi le tue configurazioni sono le seguenti:
- Devi ereditare da `Pretrainedconfig`,
- Il metodo `__init__` del tuo `Pretrainedconfig` deve accettare i kwargs,
- I `kwargs` devono essere passati alla superclass `__init__`
Lâeredità Ú importante per assicurarsi di ottenere tutte le funzionalità della libreria ð€ transformers,
mentre gli altri due vincoli derivano dal fatto che un `Pretrainedconfig` ha più campi di quelli che stai settando.
Quando ricarichi una config da un metodo `from_pretrained`, questi campi devono essere accettati dalla tua config e
poi inviati alla superclasse.
Definire un `model_type` per la tua configurazione (qua `model_type = âresnetâ`) non Ú obbligatorio, a meno che tu
non voglia registrare il modello con le classi Auto (vedi l'ultima sezione).
Una volta completato, puoi facilmente creare e salvare la tua configurazione come faresti con ogni altra configurazione
di modelli della libreria. Ecco come possiamo creare la config di un resnet50d e salvarlo:
```py
resnet50d_config = ResnetConfig(block_type="bottleneck", stem_width=32, stem_type="deep", avg_down=True)
resnet50d_config.save_pretrained("custom-resnet")
```
Questo salverà un file chiamato `config.json` all'interno della cartella `custom-resnet`. Potrai poi ricaricare la tua
config con il metodo `from_pretrained`.
```py
resnet50d_config = ResnetConfig.from_pretrained("custom-resnet")
```
Puoi anche usare qualunque altro metodo della classe [`PretrainedConfig`], come [`~PretrainedConfig.push_to_hub`]
per caricare direttamente la tua configurazione nell'hub.
## Scrivere un modello personalizzato
Ora che abbiamo la nostra configurazione ResNet, possiamo continuare a scrivere il modello. In realtà , ne scriveremo
due: uno che estrae le features nascoste da una batch di immagini (come [`BertModel`]) e uno che Ú utilizzabile per
la classificazione di immagini (come [`BertModelForSequenceClassification`]).
Come abbiamo menzionato in precedenza, scriveremo soltanto un wrapper del modello, per mantenerlo semplice ai fini di
questo esempio. L'unica cosa che dobbiamo fare prima di scrivere questa classe Ú una mappatura fra i tipi di blocco e
le vere classi dei blocchi. Successivamente il modello Ú definito tramite la configurazione, passando tutto quanto alla
classe `ResNet`.
```py
from transformers import PreTrainedModel
from timm.models.resnet import BasicBlock, Bottleneck, ResNet
from .configuration_resnet import ResnetConfig
BLOCK_MAPPING = {"basic": BasicBlock, "bottleneck": Bottleneck}
class ResnetModel(PreTrainedModel):
config_class = ResnetConfig
def __init__(self, config):
super().__init__(config)
block_layer = BLOCK_MAPPING[config.block_type]
self.model = ResNet(
block_layer,
config.layers,
num_classes=config.num_classes,
in_chans=config.input_channels,
cardinality=config.cardinality,
base_width=config.base_width,
stem_width=config.stem_width,
stem_type=config.stem_type,
avg_down=config.avg_down,
)
def forward(self, tensor):
return self.model.forward_features(tensor)
```
Per il modello che classificherà le immagini, cambiamo soltanto il metodo forward:
```py
import torch
class ResnetModelForImageClassification(PreTrainedModel):
config_class = ResnetConfig
def __init__(self, config):
super().__init__(config)
block_layer = BLOCK_MAPPING[config.block_type]
self.model = ResNet(
block_layer,
config.layers,
num_classes=config.num_classes,
in_chans=config.input_channels,
cardinality=config.cardinality,
base_width=config.base_width,
stem_width=config.stem_width,
stem_type=config.stem_type,
avg_down=config.avg_down,
)
def forward(self, tensor, labels=None):
logits = self.model(tensor)
if labels is not None:
loss = torch.nn.cross_entropy(logits, labels)
return {"loss": loss, "logits": logits}
return {"logits": logits}
```
Nota come, in entrambi i casi, ereditiamo da `PreTrainedModel` e chiamiamo l'inizializzazione della superclasse
con il metodo `config` (un po' come quando scrivi un normale `torch.nn.Module`). La riga che imposta la `config_class`
non Ú obbligatoria, a meno che tu non voglia registrare il modello con le classi Auto (vedi l'ultima sezione).
<Tip>
Se il tuo modello Ú molto simile a un modello all'interno della libreria, puoi ri-usare la stessa configurazione di quel modello.
</Tip>
Puoi fare in modo che il tuo modello restituisca in output qualunque cosa tu voglia, ma far restituire un dizionario
come abbiamo fatto per `ResnetModelForImageClassification`, con la funzione di perdita inclusa quando vengono passate le labels,
renderà il tuo modello direttamente utilizzabile all'interno della classe [`Trainer`]. Utilizzare altri formati di output va bene
se hai in progetto di utilizzare un tuo loop di allenamento, o se utilizzerai un'altra libreria per l'addestramento.
Ora che abbiamo la classe del nostro modello, creiamone uno:
```py
resnet50d = ResnetModelForImageClassification(resnet50d_config)
```
Ribadiamo, puoi usare qualunque metodo dei [`PreTrainedModel`], come [`~PreTrainedModel.save_pretrained`] o
[`~PreTrainedModel.push_to_hub`]. Utilizzeremo quest'ultimo nella prossima sezione, e vedremo come caricare i pesi del
modello assieme al codice del modello stesso. Ma prima, carichiamo alcuni pesi pre-allenati all'interno del nostro modello.
Nel tuo caso specifico, probabilmente allenerai il tuo modello sui tuoi dati. Per velocizzare in questo tutorial,
utilizzeremo la versione pre-allenata del resnet50d. Dato che il nostro modello Ú soltanto un wrapper attorno a quel modello,
sarà facile trasferirne i pesi:
```py
import timm
pretrained_model = timm.create_model("resnet50d", pretrained=True)
resnet50d.model.load_state_dict(pretrained_model.state_dict())
```
Vediamo adesso come assicurarci che quando facciamo [`~PreTrainedModel.save_pretrained`] o [`~PreTrainedModel.push_to_hub`],
il codice del modello venga salvato.
## Inviare il codice all'Hub
<Tip warning={true}>
Questa API Ú sperimentale e potrebbe avere alcuni cambiamenti nei prossimi rilasci.
</Tip>
Innanzitutto, assicurati che il tuo modello sia completamente definito in un file `.py`. Può sfruttare import relativi
ad altri file, purchÚ questi siano nella stessa directory (non supportiamo ancora sotto-moduli per questa funzionalità ).
Per questo esempio, definiremo un file `modeling_resnet.py` e un file `configuration_resnet.py` in una cartella dell'attuale
working directory chiamata `resnet_model`. Il file configuration contiene il codice per `ResnetConfig` e il file modeling
contiene il codice di `ResnetModel` e `ResnetModelForImageClassification`.
```
.
âââ resnet_model
âââ __init__.py
âââ configuration_resnet.py
âââ modeling_resnet.py
```
Il file `__init__.py` può essere vuoto, serve solo perchÚ Python capisca che `resnet_model` può essere utilizzato come un modulo.
<Tip warning={true}>
Se stai copiando i file relativi alla modellazione della libreria, dovrai sostituire tutti gli import relativi in cima al file con import del
pacchetto `transformers`.
</Tip>
Nota che puoi ri-utilizzare (o usare come sottoclassi) un modello/configurazione esistente.
Per condividere il tuo modello con la community, segui questi passi: prima importa il modello ResNet e la sua configurazione
dai nuovi file creati:
```py
from resnet_model.configuration_resnet import ResnetConfig
from resnet_model.modeling_resnet import ResnetModel, ResnetModelForImageClassification
```
DopodichÚ dovrai dire alla libreria che vuoi copiare i file con il codice di quegli oggetti quando utilizzi il metodo
`save_pretrained` e registrarli in modo corretto con una Auto classe (specialmente per i modelli). Utilizza semplicemente:
```py
ResnetConfig.register_for_auto_class()
ResnetModel.register_for_auto_class("AutoModel")
ResnetModelForImageClassification.register_for_auto_class("AutoModelForImageClassification")
```
Nota che non c'Ú bisogno di specificare una Auto classe per la configurazione (c'Ú solo una Auto classe per le configurazioni,
[`AutoConfig`], ma Ú diversa per i modelli). Il tuo modello personalizato potrebbe essere utilizzato per diverse tasks,
per cui devi specificare quale delle classi Auto Ú quella corretta per il tuo modello.
Successivamente, creiamo i modelli e la config come abbiamo fatto in precedenza:
```py
resnet50d_config = ResnetConfig(block_type="bottleneck", stem_width=32, stem_type="deep", avg_down=True)
resnet50d = ResnetModelForImageClassification(resnet50d_config)
pretrained_model = timm.create_model("resnet50d", pretrained=True)
resnet50d.model.load_state_dict(pretrained_model.state_dict())
```
Adesso, per inviare il modello all'Hub, assicurati di aver effettuato l'accesso. Lancia dal tuo terminale:
```bash
huggingface-cli login
```
O da un notebook:
```py
from huggingface_hub import notebook_login
notebook_login()
```
Potrai poi inviare il tutto sul tuo profilo (o di un'organizzazione di cui fai parte) in questo modo:
```py
resnet50d.push_to_hub("custom-resnet50d")
```
Oltre ai pesi del modello e alla configurazione in formato json, questo ha anche copiato i file `.py` modeling e
configuration all'interno della cartella `custom-resnet50d` e ha caricato i risultati sull'Hub. Puoi controllare
i risultati in questa [model repo](https://huggingface.co/sgugger/custom-resnet50d).
Puoi controllare il tutorial di condivisione [tutorial di condivisione](model_sharing) per più informazioni sul
metodo con cui inviare all'Hub.
## Usare un modello con codice personalizzato
Puoi usare ogni configurazione, modello o tokenizer con file di codice personalizzati nella sua repository
con le classi Auto e il metodo `from_pretrained`. Tutti i files e il codice caricati sull'Hub sono scansionati da malware
(fai riferimento alla documentazione [Hub security](https://huggingface.co/docs/hub/security#malware-scanning) per più informazioni),
ma dovresti comunque assicurarti dell'affidabilità del codice e dell'autore per evitare di eseguire codice dannoso sulla tua macchina.
Imposta `trust_remote_code=True` per usare un modello con codice personalizzato:
```py
from transformers import AutoModelForImageClassification
model = AutoModelForImageClassification.from_pretrained("sgugger/custom-resnet50d", trust_remote_code=True)
```
Inoltre, raccomandiamo fortemente di passare un hash del commit come `revision` per assicurarti che le autrici o gli autori del modello
non abbiano modificato il codice con alcune nuove righe dannose (a meno che non ti fidi completamente della fonte):
```py
commit_hash = "ed94a7c6247d8aedce4647f00f20de6875b5b292"
model = AutoModelForImageClassification.from_pretrained(
"sgugger/custom-resnet50d", trust_remote_code=True, revision=commit_hash
)
```
Nota che quando cerchi la storia dei commit della repo del modello sull'Hub, c'Ú un bottone con cui facilmente copiare il
commit hash di ciascun commit.
## Registrare un modello con codice personalizzato nelle classi Auto
Se stai scrivendo una libreria che estende ð€ Transformers, potresti voler estendere le classi Auto per includere il tuo modello.
Questo Ú diverso dall'inviare codice nell'Hub: gli utenti dovranno importare la tua libreria per ottenere il modello personalizzato
(anzichÚ scaricare automaticamente il modello dall'Hub).
FinchÚ il tuo file di configurazione ha un attributo `model_type` diverso dai model types esistenti, e finchÚ le tue
classi modello hanno i corretti attributi `config_class`, potrai semplicemente aggiungerli alle classi Auto come segue:
```py
from transformers import AutoConfig, AutoModel, AutoModelForImageClassification
AutoConfig.register("resnet", ResnetConfig)
AutoModel.register(ResnetConfig, ResnetModel)
AutoModelForImageClassification.register(ResnetConfig, ResnetModelForImageClassification)
```
Nota che il primo argomento utilizzato quando registri la configurazione di un modello personalizzato con [`AutoConfig`]
deve corrispondere al `model_type` della tua configurazione personalizzata, ed il primo argomento utilizzato quando
registri i tuoi modelli personalizzati in una qualunque classe Auto del modello deve corrispondere alla `config_class`
di quei modelli.
| transformers/docs/source/it/custom_models.md/0 | {
"file_path": "transformers/docs/source/it/custom_models.md",
"repo_id": "transformers",
"token_count": 5884
} | 248 |
<!--Copyright 2022 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# Pipeline per l'inferenza
La [`pipeline`] rende semplice usare qualsiasi modello dal [Model Hub](https://huggingface.co/models) per fare inferenza su diversi compiti come generazione del testo, segmentazione di immagini e classificazione di audio. Anche se non hai esperienza con una modalità specifica o non comprendi bene il codice che alimenta i modelli, Ú comunque possibile utilizzarli con l'opzione [`pipeline`]! Questa esercitazione ti insegnerà a:
* Usare una [`pipeline`] per fare inferenza.
* Usare uno specifico tokenizer o modello.
* Usare una [`pipeline`] per compiti che riguardano audio e video.
<Tip>
Dai un'occhiata alla documentazione di [`pipeline`] per una lista completa dei compiti supportati.
</Tip>
## Utilizzo della Pipeline
Nonostante ogni compito abbia una [`pipeline`] associata, Ú più semplice utilizzare l'astrazione generica della [`pipeline`] che contiene tutte quelle specifiche per ogni mansione. La [`pipeline`] carica automaticamente un modello predefinito e un tokenizer in grado di fare inferenza per il tuo compito.
1. Inizia creando una [`pipeline`] e specificando il compito su cui fare inferenza:
```py
>>> from transformers import pipeline
>>> generator = pipeline(task="text-generation")
```
2. Inserisci il testo in input nella [`pipeline`]:
```py
>>> generator(
... "Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone"
... ) # doctest: +SKIP
[{'generated_text': 'Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone, Seven for the Iron-priests at the door to the east, and thirteen for the Lord Kings at the end of the mountain'}]
```
Se hai più di un input, inseriscilo in una lista:
```py
>>> generator(
... [
... "Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone",
... "Nine for Mortal Men, doomed to die, One for the Dark Lord on his dark throne",
... ]
... ) # doctest: +SKIP
```
Qualsiasi parametro addizionale per il tuo compito può essere incluso nella [`pipeline`]. La mansione `text-generation` ha un metodo [`~generation.GenerationMixin.generate`] con diversi parametri per controllare l'output. Ad esempio, se desideri generare più di un output, utilizza il parametro `num_return_sequences`:
```py
>>> generator(
... "Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone",
... num_return_sequences=2,
... ) # doctest: +SKIP
```
### Scegliere modello e tokenizer
La [`pipeline`] accetta qualsiasi modello dal [Model Hub](https://huggingface.co/models). Ci sono tag nel Model Hub che consentono di filtrare i modelli per attività . Una volta che avrai scelto il modello appropriato, caricalo usando la corrispondente classe `AutoModelFor` e [`AutoTokenizer`]. Ad esempio, carica la classe [`AutoModelForCausalLM`] per un compito di causal language modeling:
```py
>>> from transformers import AutoTokenizer, AutoModelForCausalLM
>>> tokenizer = AutoTokenizer.from_pretrained("distilgpt2")
>>> model = AutoModelForCausalLM.from_pretrained("distilgpt2")
```
Crea una [`pipeline`] per il tuo compito, specificando il modello e il tokenizer che hai caricato:
```py
>>> from transformers import pipeline
>>> generator = pipeline(task="text-generation", model=model, tokenizer=tokenizer)
```
Inserisci il testo di input nella [`pipeline`] per generare del testo:
```py
>>> generator(
... "Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone"
... ) # doctest: +SKIP
[{'generated_text': 'Three Rings for the Elven-kings under the sky, Seven for the Dwarf-lords in their halls of stone, Seven for the Dragon-lords (for them to rule in a world ruled by their rulers, and all who live within the realm'}]
```
## Audio pipeline
La flessibilità della [`pipeline`] fa si che possa essere estesa ad attività sugli audio.
Per esempio, classifichiamo le emozioni in questo clip audio:
```py
>>> from datasets import load_dataset
>>> import torch
>>> torch.manual_seed(42) # doctest: +IGNORE_RESULT
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_demo", "clean", split="validation")
>>> audio_file = ds[0]["audio"]["path"]
```
Trova un modello per la [classificazione audio](https://huggingface.co/models?pipeline_tag=audio-classification) sul Model Hub per eseguire un compito di riconoscimento automatico delle emozioni e caricalo nella [`pipeline`]:
```py
>>> from transformers import pipeline
>>> audio_classifier = pipeline(
... task="audio-classification", model="ehcalabres/wav2vec2-lg-xlsr-en-speech-emotion-recognition"
... )
```
Inserisci il file audio nella [`pipeline`]:
```py
>>> preds = audio_classifier(audio_file)
>>> preds = [{"score": round(pred["score"], 4), "label": pred["label"]} for pred in preds]
>>> preds
[{'score': 0.1315, 'label': 'calm'}, {'score': 0.1307, 'label': 'neutral'}, {'score': 0.1274, 'label': 'sad'}, {'score': 0.1261, 'label': 'fearful'}, {'score': 0.1242, 'label': 'happy'}]
```
## Vision pipeline
Infine, usare la [`pipeline`] per le attività sulle immagini Ú praticamente la stessa cosa.
Specifica la tua attività e inserisci l'immagine nel classificatore. L'immagine può essere sia un link che un percorso sul tuo pc in locale. Per esempio, quale specie di gatto Ú raffigurata qui sotto?
```py
>>> from transformers import pipeline
>>> vision_classifier = pipeline(task="image-classification")
>>> preds = vision_classifier(
... images="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"
... )
>>> preds = [{"score": round(pred["score"], 4), "label": pred["label"]} for pred in preds]
>>> preds
[{'score': 0.4335, 'label': 'lynx, catamount'}, {'score': 0.0348, 'label': 'cougar, puma, catamount, mountain lion, painter, panther, Felis concolor'}, {'score': 0.0324, 'label': 'snow leopard, ounce, Panthera uncia'}, {'score': 0.0239, 'label': 'Egyptian cat'}, {'score': 0.0229, 'label': 'tiger cat'}]
```
| transformers/docs/source/it/pipeline_tutorial.md/0 | {
"file_path": "transformers/docs/source/it/pipeline_tutorial.md",
"repo_id": "transformers",
"token_count": 2390
} | 249 |
<!--Copyright 2023 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# Templates for Chat Models
## Introduction
LLMïŒLanguage ModelïŒã®ãŸããŸãäžè¬çãªäœ¿çšäºäŸã®1ã€ã¯ããã£ãããã§ãã
ãã£ããã®ã³ã³ããã¹ãã§ã¯ãéåžžã®èšèªã¢ãã«ã®ããã«åäžã®ããã¹ãã¹ããªã³ã°ãç¶ç¶ããã®ã§ã¯ãªããã¢ãã«ã¯1ã€ä»¥äžã®ãã¡ãã»ãŒãžããããªãäŒè©±ãç¶ç¶ããŸãã
åã¡ãã»ãŒãžã«ã¯ãããŒã«ããšã¡ãã»ãŒãžããã¹ããå«ãŸããŸãã
æãäžè¬çã«ããããã®ããŒã«ã¯ãŠãŒã¶ãŒããã®ã¡ãã»ãŒãžã«ã¯ããŠãŒã¶ãŒããã¢ãã«ããã®ã¡ãã»ãŒãžã«ã¯ãã¢ã·ã¹ã¿ã³ãããå²ãåœãŠãããŸãã
äžéšã®ã¢ãã«ã¯ãã·ã¹ãã ãããŒã«ããµããŒãããŠããŸãã
ã·ã¹ãã ã¡ãã»ãŒãžã¯éåžžäŒè©±ã®éå§æã«éä¿¡ãããã¢ãã«ã®åäœæ¹æ³ã«é¢ããæ瀺ãå«ãŸããŸãã
ãã¹ãŠã®èšèªã¢ãã«ããã£ããçšã«åŸ®èª¿æŽãããã¢ãã«ãå«ããã¹ãŠã®ã¢ãã«ã¯ãããŒã¯ã³ã®ãªãã¢ã·ãŒã±ã³ã¹ã§åäœããããŒã«ã«ç¹æã®ç¹å¥ãªåŠçãæã¡ãŸããã
ã€ãŸããããŒã«æ
å ±ã¯éåžžãã¡ãã»ãŒãžéã«å¶åŸ¡ããŒã¯ã³ãè¿œå ããŠæ³šå
¥ãããã¡ãã»ãŒãžã®å¢çãšé¢é£ããããŒã«ã瀺ãããšã§æäŸãããŸãã
æ®å¿µãªãããããŒã¯ã³ã®äœ¿çšæ¹æ³ã«ã€ããŠã¯ïŒãŸã ïŒïŒæšæºãååšãããç°ãªãã¢ãã«ã¯ãã£ããçšã®ãã©ãŒããããå¶åŸ¡ããŒã¯ã³ã倧ããç°ãªã圢åŒã§ãã¬ãŒãã³ã°ãããŠããŸãã
ããã¯ãŠãŒã¶ãŒã«ãšã£ãŠå®éã®åé¡ã«ãªãå¯èœæ§ããããŸããæ£ãããã©ãŒãããã䜿çšããªããšãã¢ãã«ã¯å
¥åã«æ··ä¹±ããããã©ãŒãã³ã¹ãæ¬æ¥ãããé¥ãã«äœäžããŸãã
ãããããã£ãããã³ãã¬ãŒããã解決ããããšããåé¡ã§ãã
ãã£ããäŒè©±ã¯éåžžãåèŸæžããããŒã«ããšãã³ã³ãã³ããã®ããŒãå«ã¿ãåäžã®ãã£ããã¡ãã»ãŒãžãè¡šããªã¹ããšããŠè¡šçŸãããŸãã
ãã£ãããã³ãã¬ãŒãã¯ãæå®ãããã¢ãã«ã®äŒè©±ãåäžã®ããŒã¯ã³åå¯èœãªã·ãŒã±ã³ã¹ã«ã©ã®ããã«ãã©ãŒãããããããæå®ããJinjaãã³ãã¬ãŒããå«ãæååã§ãã
ããŒã¯ãã€ã¶ãšãã®æ
å ±ãä¿åããããšã«ãããã¢ãã«ãæåŸ
ãã圢åŒã®å
¥åããŒã¿ãååŸã§ããããã«ãªããŸãã
ãã£ããã`BlenderBot` ã¢ãã«ã䜿çšããäŸã瀺ããŠå
·äœçã«ããŸãããã`BlenderBot` ã®ããã©ã«ããã³ãã¬ãŒãã¯éåžžã«ã·ã³ãã«ã§ãã»ãšãã©ã察話ã®ã©ãŠã³ãéã«ç©ºçœãè¿œå ããã ãã§ãã
```python
>>> from transformers import AutoTokenizer
>>> tokenizer = AutoTokenizer.from_pretrained("facebook/blenderbot-400M-distill")
>>> chat = [
... {"role": "user", "content": "Hello, how are you?"},
... {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
... {"role": "user", "content": "I'd like to show off how chat templating works!"},
... ]
>>> tokenizer.apply_chat_template(chat, tokenize=False)
" Hello, how are you? I'm doing great. How can I help you today? I'd like to show off how chat templating works!</s>"
```
æå®ãããéãããã£ããå
šäœãåäžã®æååã«ãŸãšããããŠããŸããããã©ã«ãã®èšå®ã§ãããtokenize=Trueãã䜿çšãããšã
ãã®æååãããŒã¯ã³åãããŸããããããããè€éãªãã³ãã¬ãŒããå®éã«ã©ã®ããã«æ©èœãããã確èªããããã«ã
ãmeta-llama/Llama-2-7b-chat-hfãã¢ãã«ã䜿çšããŠã¿ãŸãããããã ãããã®ã¢ãã«ã¯ã²ãŒãä»ãã¢ã¯ã»ã¹ãæã£ãŠããã
ãã®ã³ãŒããå®è¡ããå Žåã¯[ãªããžããªã§ã¢ã¯ã»ã¹ããªã¯ãšã¹ã](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf)ããå¿
èŠããããŸãã
```python
>> from transformers import AutoTokenizer
>> tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-chat-hf")
>> chat = [
... {"role": "user", "content": "Hello, how are you?"},
... {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
... {"role": "user", "content": "I'd like to show off how chat templating works!"},
... ]
>> tokenizer.use_default_system_prompt = False
>> tokenizer.apply_chat_template(chat, tokenize=False)
"<s>[INST] Hello, how are you? [/INST] I'm doing great. How can I help you today? </s><s>[INST] I'd like to show off how chat templating works! [/INST]"
```
ä»åãããŒã¯ãã€ã¶ã¯å¶åŸ¡ããŒã¯ã³ [INST] ãš [/INST] ãè¿œå ããŸããããããã¯ãŠãŒã¶ãŒã¡ãã»ãŒãžã®éå§ãšçµäºã瀺ãããã®ãã®ã§ãïŒãã ããã¢ã·ã¹ã¿ã³ãã¡ãã»ãŒãžã«ã¯é©çšãããŸããïŒïŒ
## How do chat templates work?
ã¢ãã«ã®ãã£ãããã³ãã¬ãŒãã¯ã`tokenizer.chat_template`å±æ§ã«æ ŒçŽãããŠããŸãããã£ãããã³ãã¬ãŒããèšå®ãããŠããªãå Žåããã®ã¢ãã«ã¯ã©ã¹ã®ããã©ã«ããã³ãã¬ãŒãã代ããã«äœ¿çšãããŸãã`BlenderBot`ã®ãã³ãã¬ãŒããèŠãŠã¿ãŸããã:
```python
>>> from transformers import AutoTokenizer
>>> tokenizer = AutoTokenizer.from_pretrained("facebook/blenderbot-400M-distill")
>>> tokenizer.default_chat_template
"{% for message in messages %}{% if message['role'] == 'user' %}{{ ' ' }}{% endif %}{{ message['content'] }}{% if not loop.last %}{{ ' ' }}{% endif %}{% endfor %}{{ eos_token }}"
```
ããã¯å°ãæå§çã§ãããå¯èªæ§ãé«ããããã«ãæ°ããè¡ãšã€ã³ãã³ããè¿œå ããŸãããã
åãããã¯ã®çŽåã®ç©ºçœãšããããã¯ã®çŽåŸã®æåã®æ¹è¡ã¯ãããã©ã«ãã§Jinjaã® `trim_blocks` ããã³ `lstrip_blocks` ãã©ã°ã䜿çšããŠåé€ããŸãã
ããã«ãããã€ã³ãã³ããšæ¹è¡ãå«ããã³ãã¬ãŒããæžããŠãæ£åžžã«æ©èœããããšãã§ããŸãã
```
{% for message in messages %}
{% if message['role'] == 'user' %}
{{ ' ' }}
{% endif %}
{{ message['content'] }}
{% if not loop.last %}
{{ ' ' }}
{% endif %}
{% endfor %}
{{ eos_token }}
```
ãããåããŠèŠãæ¹ãžãããã¯[Jinjaãã³ãã¬ãŒã](https://jinja.palletsprojects.com/en/3.1.x/templates/)ã§ãã
Jinjaã¯ããã¹ããçæããããã®ã·ã³ãã«ãªã³ãŒããèšè¿°ã§ãããã³ãã¬ãŒãèšèªã§ããå€ãã®ç¹ã§ãã³ãŒããš
æ§æã¯Pythonã«äŒŒãŠããŸããçŽç²ãªPythonã§ã¯ããã®ãã³ãã¬ãŒãã¯æ¬¡ã®ããã«ãªãã§ãããïŒ
```python
for idx, message in enumerate(messages):
if message['role'] == 'user':
print(' ')
print(message['content'])
if not idx == len(messages) - 1: # Check for the last message in the conversation
print(' ')
print(eos_token)
```
å®éã«ããã®ãã³ãã¬ãŒãã¯æ¬¡ã®3ã€ã®ããšãè¡ããŸãïŒ
1. åã¡ãã»ãŒãžã«å¯ŸããŠãã¡ãã»ãŒãžããŠãŒã¶ãŒã¡ãã»ãŒãžã§ããå Žåãããã®åã«ç©ºçœãè¿œå ãããã以å€ã®å Žåã¯äœã衚瀺ããŸããã
2. ã¡ãã»ãŒãžã®å
容ãè¿œå ããŸãã
3. ã¡ãã»ãŒãžãæåŸã®ã¡ãã»ãŒãžã§ãªãå Žåããã®åŸã«2ã€ã®ã¹ããŒã¹ãè¿œå ããŸããæåŸã®ã¡ãã»ãŒãžã®åŸã«ã¯EOSããŒã¯ã³ã衚瀺ããŸãã
ããã¯éåžžã«ã·ã³ãã«ãªãã³ãã¬ãŒãã§ããå¶åŸ¡ããŒã¯ã³ãè¿œå ããªãããã¢ãã«ã«å¯Ÿããæ瀺ãäŒããäžè¬çãªæ¹æ³ã§ãããã·ã¹ãã ãã¡ãã»ãŒãžããµããŒãããŠããŸããã
ãã ããJinjaã¯ãããã®ããšãè¡ãããã®å€ãã®æè»æ§ãæäŸããŠããŸãïŒ
LLaMAããã©ãŒãããããæ¹æ³ã«é¡äŒŒããå
¥åããã©ãŒãããããããã®Jinjaãã³ãã¬ãŒããèŠãŠã¿ãŸããã
ïŒå®éã®LLaMAãã³ãã¬ãŒãã¯ããã©ã«ãã®ã·ã¹ãã ã¡ãã»ãŒãžã®åŠçããäžè¬çãªã·ã¹ãã ã¡ãã»ãŒãžã®åŠçãè¥å¹²ç°ãªãããã
å®éã®ã³ãŒãã§ã¯ãã®ãã³ãã¬ãŒãã䜿çšããªãã§ãã ããïŒïŒ
```
{% for message in messages %}
{% if message['role'] == 'user' %}
{{ bos_token + '[INST] ' + message['content'] + ' [/INST]' }}
{% elif message['role'] == 'system' %}
{{ '<<SYS>>\\n' + message['content'] + '\\n<</SYS>>\\n\\n' }}
{% elif message['role'] == 'assistant' %}
{{ ' ' + message['content'] + ' ' + eos_token }}
{% endif %}
{% endfor %}
```
é¡ããã°ãå°ãèŠã€ããŠããã ããã°ããã®ãã³ãã¬ãŒããäœãè¡ã£ãŠããããããããããããŸããã
ãã®ãã³ãã¬ãŒãã¯ãåã¡ãã»ãŒãžã®ã圹å²ãã«åºã¥ããŠç¹å®ã®ããŒã¯ã³ãè¿œå ããŸãããããã®ããŒã¯ã³ã¯ãã¡ãã»ãŒãžãéä¿¡ãã人ãè¡šããã®ã§ãã
ãŠãŒã¶ãŒãã¢ã·ã¹ã¿ã³ããããã³ã·ã¹ãã ã¡ãã»ãŒãžã¯ãããããå«ãŸããããŒã¯ã³ã«ãã£ãŠã¢ãã«ã«ãã£ãŠæ確ã«åºå¥ãããŸãã
## How do I create a chat template?
ç°¡åã§ããåçŽã«Jinjaãã³ãã¬ãŒããæžããŠã`tokenizer.chat_template`ãèšå®ããŸãã
ä»ã®ã¢ãã«ããæ¢åã®ãã³ãã¬ãŒããå§ç¹ã«ããŠãå¿
èŠã«å¿ããŠç·šéãããšäŸ¿å©ãããããŸããïŒ
äŸãã°ãäžèšã®LLaMAãã³ãã¬ãŒããåã£ãŠãã¢ã·ã¹ã¿ã³ãã¡ãã»ãŒãžã«"[ASST]"ãš"[/ASST]"ãè¿œå ã§ããŸãã
```
{% for message in messages %}
{% if message['role'] == 'user' %}
{{ bos_token + '[INST] ' + message['content'].strip() + ' [/INST]' }}
{% elif message['role'] == 'system' %}
{{ '<<SYS>>\\n' + message['content'].strip() + '\\n<</SYS>>\\n\\n' }}
{% elif message['role'] == 'assistant' %}
{{ '[ASST] ' + message['content'] + ' [/ASST]' + eos_token }}
{% endif %}
{% endfor %}
```
次ã«ãåã«`tokenizer.chat_template`å±æ§ãèšå®ããŠãã ããã
次åã[`~PreTrainedTokenizer.apply_chat_template`]ã䜿çšããéã«ãæ°ãããã³ãã¬ãŒãã䜿çšãããŸãïŒ
ãã®å±æ§ã¯`tokenizer_config.json`ãã¡ã€ã«ã«ä¿åãããããã[`~utils.PushToHubMixin.push_to_hub`]ã䜿çšããŠ
æ°ãããã³ãã¬ãŒããHubã«ã¢ããããŒãããã¿ããªãæ£ãããã³ãã¬ãŒãã䜿çšããŠããããšã確èªã§ããŸãïŒ
```python
template = tokenizer.chat_template
template = template.replace("SYS", "SYSTEM") # Change the system token
tokenizer.chat_template = template # Set the new template
tokenizer.push_to_hub("model_name") # Upload your new template to the Hub!
```
[`~PreTrainedTokenizer.apply_chat_template`] ã¡ãœããã¯ãããªãã®ãã£ãããã³ãã¬ãŒãã䜿çšããããã« [`ConversationalPipeline`] ã¯ã©ã¹ã«ãã£ãŠåŒã³åºãããŸãã
ãããã£ãŠãæ£ãããã£ãããã³ãã¬ãŒããèšå®ãããšãããªãã®ã¢ãã«ã¯èªåçã« [`ConversationalPipeline`] ãšäºææ§ãããããã«ãªããŸãã
## What are "default" templates?
ãã£ãããã³ãã¬ãŒãã®å°å
¥åã«ããã£ããã®åŠçã¯ã¢ãã«ã¯ã©ã¹ã¬ãã«ã§ããŒãã³ãŒããããŠããŸããã
åŸæ¹äºææ§ã®ããã«ããã®ã¯ã©ã¹åºæã®åŠçãããã©ã«ããã³ãã¬ãŒããšããŠä¿æããã¯ã©ã¹ã¬ãã«ã§èšå®ãããŠããŸãã
ã¢ãã«ã«ãã£ãããã³ãã¬ãŒããèšå®ãããŠããªãå Žåããã ãã¢ãã«ã¯ã©ã¹ã®ããã©ã«ããã³ãã¬ãŒããããå Žåã
`ConversationalPipeline`ã¯ã©ã¹ã`apply_chat_template`ãªã©ã®ã¡ãœããã¯ã¯ã©ã¹ãã³ãã¬ãŒãã䜿çšããŸãã
ããŒã¯ãã€ã¶ã®ããã©ã«ãã®ãã£ãããã³ãã¬ãŒãã確èªããã«ã¯ã`tokenizer.default_chat_template`å±æ§ããã§ãã¯ããŠãã ããã
ããã¯ãåŸæ¹äºææ§ã®ããã«çŽç²ã«è¡ã£ãŠããããšã§ãæ¢åã®ã¯ãŒã¯ãããŒãå£ããªãããã«ããŠããŸãã
ã¢ãã«ã«ãšã£ãŠã¯ã©ã¹ãã³ãã¬ãŒããé©åã§ããå Žåã§ããããã©ã«ããã³ãã¬ãŒãããªãŒããŒã©ã€ãããŠ
`chat_template`å±æ§ãæ瀺çã«èšå®ããããšã匷ããå§ãããŸããããã«ããããŠãŒã¶ãŒã«ãšã£ãŠ
ã¢ãã«ããã£ããçšã«æ£ããæ§æãããŠããããšãæ確ã«ãªããããã©ã«ããã³ãã¬ãŒããå€æŽããããå»æ¢ãããå Žåã«åããããšãã§ããŸãã
## What template should I use?
ãã§ã«ãã£ããã®ãã¬ãŒãã³ã°ãåããã¢ãã«ã®ãã³ãã¬ãŒããèšå®ããå Žåããã³ãã¬ãŒãããã¬ãŒãã³ã°äžã«ã¢ãã«ãèŠãã¡ãã»ãŒãžã®ãã©ãŒããããšãŸã£ããäžèŽããããšã確èªããå¿
èŠããããŸãã
ããã§ãªãå Žåãæ§èœã®äœäžãçµéšããå¯èœæ§ãé«ãã§ããããã¯ã¢ãã«ãããã«ãã¬ãŒãã³ã°ããŠããå Žåã§ãåæ§ã§ã - ãã£ããããŒã¯ã³ãäžå®ã«ä¿ã€ãšãããããæé«ã®æ§èœãåŸãããŸãã
ããã¯ããŒã¯ã³åãšéåžžã«é¡äŒŒããŠãããéåžžã¯ãã¬ãŒãã³ã°äžã«äœ¿çšãããããŒã¯ã³åãšæ£ç¢ºã«äžèŽããå Žåã«ãæšè«ãŸãã¯ãã¡ã€ã³ãã¥ãŒãã³ã°ã®éã«æè¯ã®æ§èœãåŸãããŸãã
äžæ¹ããŒãããã¢ãã«ããã¬ãŒãã³ã°ãããããã£ããã®ããã«ããŒã¹èšèªã¢ãã«ããã¡ã€ã³ãã¥ãŒãã³ã°ããå Žåãé©åãªãã³ãã¬ãŒããéžæããèªç±åºŠããããŸãã
LLMïŒLanguage ModelïŒã¯ããŸããŸãªå
¥å圢åŒãåŠçã§ããã»ã©ã¹ããŒãã§ããã¯ã©ã¹åºæã®ãã³ãã¬ãŒãããªãã¢ãã«çšã®ããã©ã«ããã³ãã¬ãŒãã¯ãäžè¬çãªãŠãŒã¹ã±ãŒã¹ã«å¯ŸããŠè¯ãæè»ãªéžæè¢ã§ãã
ããã¯ã[ChatMLãã©ãŒããã](https://github.com/openai/openai-python/blob/main/chatml.md)ã«åŸã£ããã®ã§ãå€ãã®ãŠãŒã¹ã±ãŒã¹ã«é©ããŠããŸãã次ã®ããã«ãªããŸãïŒ
```
{% for message in messages %}
{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}
{% endfor %}
```
If you like this one, here it is in one-liner form, ready to copy into your code:
```
tokenizer.chat_template = "{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}"
```
ãã®ãã³ãã¬ãŒãã¯ãåã¡ãã»ãŒãžãã``ãããŒã¯ã³ã§å²ã¿ã圹å²ãæååãšããŠåçŽã«èšè¿°ããŸãã
ããã«ããããã¬ãŒãã³ã°ã§äœ¿çšãã圹å²ã«å¯Ÿããæè»æ§ãåŸãããŸããåºåã¯ä»¥äžã®ããã«ãªããŸãïŒ
```
<|im_start|>system
You are a helpful chatbot that will do its best not to say anything so stupid that people tweet about it.<|im_end|>
<|im_start|>user
How are you?<|im_end|>
<|im_start|>assistant
I'm doing great!<|im_end|>
```
ããŠãŒã¶ãŒãããã·ã¹ãã ããããã³ãã¢ã·ã¹ã¿ã³ããã®åœ¹å²ã¯ããã£ããã®æšæºã§ãã
ç¹ã«ã[`ConversationalPipeline`]ãšã®é£æºãã¹ã ãŒãºã«è¡ãå Žåã«ã¯ããããã®åœ¹å²ã䜿çšããããšããå§ãããŸãããã ãããããã®åœ¹å²ã«å¶çŽã¯ãããŸããããã³ãã¬ãŒãã¯éåžžã«æè»ã§ãä»»æã®æååã圹å²ãšããŠäœ¿çšã§ããŸãã
## I want to use chat templates! How should I get started?
ãã£ããã¢ãã«ãæã£ãŠããå Žåããã®ã¢ãã«ã®`tokenizer.chat_template`å±æ§ãèšå®ãã[`~PreTrainedTokenizer.apply_chat_template`]ã䜿çšããŠãã¹ãããå¿
èŠããããŸãã
ããã¯ã¢ãã«ã®ææè
ã§ãªãå Žåã§ãé©çšãããŸããã¢ãã«ã®ãªããžããªã空ã®ãã£ãããã³ãã¬ãŒãã䜿çšããŠããå ŽåããŸãã¯ããã©ã«ãã®ã¯ã©ã¹ãã³ãã¬ãŒãã䜿çšããŠããå Žåã§ãã
ãã®å±æ§ãé©åã«èšå®ã§ããããã«[ãã«ãªã¯ãšã¹ã](https://huggingface.co/docs/hub/repositories-pull-requests-discussions)ãéããŠãã ããã
äžåºŠå±æ§ãèšå®ãããã°ãããã§å®äºã§ãïŒ `tokenizer.apply_chat_template`ã¯ããã®ã¢ãã«ã«å¯ŸããŠæ£ããåäœããããã«ãªããŸããããã¯ã
`ConversationalPipeline`ãªã©ã®å Žæã§ãèªåçã«ãµããŒããããŸãã
ã¢ãã«ããã®å±æ§ãæã€ããšã確èªããããšã§ããªãŒãã³ãœãŒã¹ã¢ãã«ã®å
šã³ãã¥ããã£ããã®ãã«ãã¯ãŒã䜿çšã§ããããã«ãªããŸãã
ãã©ãŒãããã®äžäžèŽã¯ãã®åéã«æ©ã¿ç¶ããããã©ãŒãã³ã¹ã«é»ã£ãŠåœ±é¿ãäžããŠããŸããããããçµããããæãæ¥ãŸããïŒ
| transformers/docs/source/ja/chat_templating.md/0 | {
"file_path": "transformers/docs/source/ja/chat_templating.md",
"repo_id": "transformers",
"token_count": 7022
} | 250 |
<!--Copyright 2020 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# BARThez
## Overview
BARThez ã¢ãã«ã¯ãMoussa Kamal EddineãAntoine J.-P ã«ãã£ãŠ [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) ã§ææ¡ãããŸããããã£ã¯ã·ãšããã«ãªã¹ã»ãŽã¡ãžã«ãžã£ã³ãã¹ã10æ23æ¥ã
2020幎ã
è«æã®èŠçŽ:
*åž°çŽç転移åŠç¿ã¯ãèªå·±æåž«ããåŠç¿ã«ãã£ãŠå¯èœã«ãªããèªç¶èšèªåŠçå
šäœãå®è¡ããŸãã
(NLP) åéã¯ãBERT ã BART ãªã©ã®ã¢ãã«ã«ãããç¡æ°ã®èªç¶èšèªã«æ°ããªæå
端æè¡ã確ç«ããåµãå·»ãèµ·ãããŠããŸãã
ã¿ã¹ã¯ãç解ããããšãããã€ãã®æ³šç®ãã¹ãäŸå€ã¯ãããŸãããå©çšå¯èœãªã¢ãã«ãšç 究ã®ã»ãšãã©ã¯ã
è±èªã察象ã«å®æœãããŸããããã®äœåã§ã¯ããã©ã³ã¹èªçšã®æåã® BART ã¢ãã«ã§ãã BARTez ã玹ä»ããŸãã
ïŒæã
ã®ç¥ãéãã«ïŒã BARThez ã¯ãéå»ã®ç 究ããåŸãéåžžã«å€§èŠæš¡ãªåäžèšèªãã©ã³ã¹èªã³ãŒãã¹ã§äºåãã¬ãŒãã³ã°ãããŸãã
BART ã®æåã¹ããŒã ã«åãããŠèª¿æŽããŸãããæ¢åã® BERT ããŒã¹ã®ãã©ã³ã¹èªã¢ãã«ãšã¯ç°ãªãã
CamemBERT ãš FlauBERTãBARThez ã¯ããšã³ã³ãŒãã ãã§ãªãã
ãã®ãã³ãŒãã¯äºåãã¬ãŒãã³ã°ãããŠããŸãã FLUE ãã³ãããŒã¯ããã®èå¥ã¿ã¹ã¯ã«å ããŠãBARThez ãæ°ããè©äŸ¡ã«åºã¥ããŠè©äŸ¡ããŸãã
ãã®è«æãšãšãã«ãªãªãŒã¹ããèŠçŽããŒã¿ã»ãããOrangeSumããŸãããã§ã«è¡ãããŠããäºåãã¬ãŒãã³ã°ãç¶ç¶ããŸãã
BARTHez ã®ã³ãŒãã¹äžã§å€èšèª BART ãäºåèšç·ŽããçµæãšããŠåŸãããã¢ãã« (mBARTHez ãšåŒã¶) ã次ã®ããšã瀺ããŸãã
ããã©ã® BARThez ã倧å¹
ã«åŒ·åããCamemBERT ã FlauBERT ãšåçããããäžåããŸãã*
ãã®ã¢ãã«ã¯ [moussakam](https://huggingface.co/moussakam) ã«ãã£ãŠå¯çš¿ãããŸãããèè
ã®ã³ãŒãã¯[ãã](https://github.com/moussaKam/BARThez)ã«ãããŸãã
<Tip>
BARThez ã®å®è£
ã¯ãããŒã¯ã³åãé€ã㊠BART ãšåãã§ãã詳现ã«ã€ããŠã¯ã[BART ããã¥ã¡ã³ã](bart) ãåç
§ããŠãã ããã
æ§æã¯ã©ã¹ãšãã®ãã©ã¡ãŒã¿ã BARThez åºæã®ããŒã¯ãã€ã¶ãŒã«ã€ããŠã¯ä»¥äžã«èšèŒãããŠããŸãã
</Tip>
### Resources
- BARThez ã¯ãBART ãšåæ§ã®æ¹æ³ã§ã·ãŒã±ã³ã¹éã®ã¿ã¹ã¯ã埮調æŽã§ããŸãã以äžã確èªããŠãã ããã
[examples/pytorch/summarization/](https://github.com/huggingface/transformers/tree/main/examples/pytorch/summarization/README.md)ã
## BarthezTokenizer
[[autodoc]] BarthezTokenizer
## BarthezTokenizerFast
[[autodoc]] BarthezTokenizerFast
| transformers/docs/source/ja/model_doc/barthez.md/0 | {
"file_path": "transformers/docs/source/ja/model_doc/barthez.md",
"repo_id": "transformers",
"token_count": 1461
} | 251 |
<!--Copyright 2020 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# BORT
<Tip warning={true}>
ãã®ã¢ãã«ã¯ã¡ã³ããã³ã¹ ã¢ãŒãã®ã¿ã§ãããã³ãŒããå€æŽããæ°ãã PR ã¯åãä»ããããŸããã
ãã®ã¢ãã«ã®å®è¡äžã«åé¡ãçºçããå Žåã¯ããã®ã¢ãã«ããµããŒãããŠããæåŸã®ããŒãžã§ã³ (v4.30.0) ãåã€ã³ã¹ããŒã«ããŠãã ããã
ãããè¡ãã«ã¯ãã³ãã³ã `pip install -U Transformers==4.30.0` ãå®è¡ããŸãã
</Tip>
## Overview
BORT ã¢ãã«ã¯ã[Optimal Subarchitecture Extraction for BERT](https://arxiv.org/abs/2010.10499) ã§ææ¡ãããŸããã
Adrian de Wynter and Daniel J. Perry.ããã¯ãBERT ã®ã¢ãŒããã¯ã㣠ãã©ã¡ãŒã¿ã®æé©ãªãµãã»ããã§ãã
èè
ã¯ããã«ãããšåŒãã§ããŸãã
è«æã®èŠçŽã¯æ¬¡ã®ãšããã§ãã
*Devlin ããã BERT ã¢ãŒããã¯ãã£ã®ã¢ãŒããã¯ã㣠ãã©ã¡ãŒã¿ã®æé©ãªãµãã»ãããæœåºããŸãã (2018)
ãã¥ãŒã©ã« ã¢ãŒããã¯ãã£æ€çŽ¢ã®ã¢ã«ãŽãªãºã ã«ãããæè¿ã®ç»æçãªæè¡ãé©çšããŸãããã®æé©ãªãµãã»ããã次ã®ããã«åŒã³ãŸãã
"Bort" ã¯æããã«å°ãããæå¹ (ã€ãŸããåã蟌ã¿å±€ãèæ
®ããªã) ãµã€ãºã¯ 5.5% ã§ãã
ãªãªãžãã«ã® BERT 倧èŠæš¡ã¢ãŒããã¯ãã£ãããã³ããã ãµã€ãºã® 16%ã Bort 㯠288 GPU æéã§äºåãã¬ãŒãã³ã°ããããšãã§ããŸãã
æé«ããã©ãŒãã³ã¹ã® BERT ãã©ã¡ããªã㯠ã¢ãŒããã¯ã㣠ããªã¢ã³ãã§ãã RoBERTa-large ã®äºåãã¬ãŒãã³ã°ã«å¿
èŠãªæéã® 1.2%
(Liu et al., 2019)ãåããã·ã³ã§ BERT-large ããã¬ãŒãã³ã°ããã®ã«å¿
èŠãª GPU æéã®äžçèšé²ã®çŽ 33%
ããŒããŠã§ã¢ããŸããCPU äžã§ 7.9 åé«éã§ããã ãã§ãªããä»ã®å§çž®ããŒãžã§ã³ãããããã©ãŒãã³ã¹ãåªããŠããŸãã
ã¢ãŒããã¯ãã£ãããã³äžéšã®éå§çž®ããªã¢ã³ã: 0.3% ïœ 31% ã®ããã©ãŒãã³ã¹åäžãåŸãããŸãã
BERT-large ã«é¢ããŠãè€æ°ã®å
¬éèªç¶èšèªç解 (NLU) ãã³ãããŒã¯ã«ããã絶察çãªè©äŸ¡ã*
ãã®ã¢ãã«ã¯ [stefan-it](https://huggingface.co/stefan-it) ã«ãã£ãŠæäŸãããŸãããå
ã®ã³ãŒãã¯[ãã](https://github.com/alexa/bort/)ã«ãããŸãã
## Usage tips
- BORT ã®ã¢ãã« ã¢ãŒããã¯ãã£ã¯ BERT ã«åºã¥ããŠããŸãã詳现ã«ã€ããŠã¯ã[BERT ã®ããã¥ã¡ã³ã ããŒãž](bert) ãåç
§ããŠãã ããã
ã¢ãã«ã® API ãªãã¡ã¬ã³ã¹ãšäœ¿çšäŸã
- BORT 㯠BERT ããŒã¯ãã€ã¶ãŒã®ä»£ããã« RoBERTa ããŒã¯ãã€ã¶ãŒã䜿çšããŸããããŒã¯ãã€ã¶ãŒã® API ãªãã¡ã¬ã³ã¹ãšäœ¿çšäŸã«ã€ããŠã¯ã[RoBERTa ã®ããã¥ã¡ã³ã ããŒãž](roberta) ãåç
§ããŠãã ããã
- BORT ã«ã¯ã [Agora](https://adewynter.github.io/notes/bort_algorithms_and_applications.html#fine-tuning-with-algebraic-topology) ãšåŒã°ããç¹å®ã®åŸ®èª¿æŽã¢ã«ãŽãªãºã ãå¿
èŠã§ãã
æ®å¿µãªãããŸã ãªãŒãã³ãœãŒã¹åãããŠããŸããã誰ããå®è£
ããããšãããšãã³ãã¥ããã£ã«ãšã£ãŠéåžžã«åœ¹ç«ã¡ãŸãã
BORT ã®åŸ®èª¿æŽãæ©èœãããããã®ã¢ã«ãŽãªãºã ã | transformers/docs/source/ja/model_doc/bort.md/0 | {
"file_path": "transformers/docs/source/ja/model_doc/bort.md",
"repo_id": "transformers",
"token_count": 1598
} | 252 |
<!--Copyright 2023 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
â ïž Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
rendered properly in your Markdown viewer.
-->
# ConvNeXt V2
## Overview
ConvNeXt V2 ã¢ãã«ã¯ãSanghyun WooãShobhik DebnathãRonghang HuãXinlei ChenãZhuang Liu, In So Kweon, Saining Xie. ã«ãã£ãŠ [ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders](https://arxiv.org/abs/2301.00808) ã§ææ¡ãããŸããã
ConvNeXt V2 ã¯ãVision Transformers ã®èšèšããã€ã³ã¹ãã¬ãŒã·ã§ã³ãåŸãçŽç²ãªç³ã¿èŸŒã¿ã¢ãã« (ConvNet) ã§ããã[ConvNeXT](convnext) ã®åŸç¶ã§ãã
è«æã®èŠçŽã¯æ¬¡ã®ãšããã§ãã
*ã¢ãŒããã¯ãã£ã®æ¹åãšè¡šçŸåŠç¿ãã¬ãŒã ã¯ãŒã¯ã®æ¹åã«ãããèŠèŠèªèã®åé㯠2020 幎代åé ã«æ¥éãªè¿ä»£åãšããã©ãŒãã³ã¹ã®åäžãå®çŸããŸãããããšãã°ãConvNeXt ã«ä»£è¡šãããææ°ã® ConvNet ã¯ãããŸããŸãªã·ããªãªã§åŒ·åãªããã©ãŒãã³ã¹ãå®èšŒããŠããŸãããããã®ã¢ãã«ã¯ããšããš ImageNet ã©ãã«ã䜿çšããæåž«ããåŠç¿çšã«èšèšãããŸãããããã¹ã¯ ãªãŒããšã³ã³ãŒã㌠(MAE) ãªã©ã®èªå·±æåž«ããåŠç¿ææ³ãããæœåšçã«æ©æµãåããããšãã§ããŸãããã ããããã 2 ã€ã®ã¢ãããŒããåçŽã«çµã¿åããããšãããã©ãŒãã³ã¹ãæšæºä»¥äžã«ãªãããšãããããŸããããã®è«æã§ã¯ãå®å
šç³ã¿èŸŒã¿ãã¹ã¯ ãªãŒããšã³ã³ãŒã ãã¬ãŒã ã¯ãŒã¯ãšããã£ãã«éã®æ©èœç«¶åã匷åããããã« ConvNeXt ã¢ãŒããã¯ãã£ã«è¿œå ã§ããæ°ãã Global Response Normalization (GRN) å±€ãææ¡ããŸãããã®èªå·±æåž«ããåŠç¿ææ³ãšã¢ãŒããã¯ãã£ã®æ¹åã®å
±åèšèšã«ãããConvNeXt V2 ãšåŒã°ããæ°ããã¢ãã« ãã¡ããªãèªçããŸãããããã«ãããImageNet åé¡ãCOCO æ€åºãADE20K ã»ã°ã¡ã³ããŒã·ã§ã³ãªã©ã®ããŸããŸãªèªèãã³ãããŒã¯ã«ãããçŽç²ãª ConvNet ã®ããã©ãŒãã³ã¹ã倧å¹
ã«åäžããŸãããŸããImageNet ã§ããã 1 ã®ç²ŸåºŠ 76.7% ãèªãå¹çç㪠370 äžãã©ã¡ãŒã¿ã® Atto ã¢ãã«ãããæå
端㮠88.9% ãéæãã 650M Huge ã¢ãã«ãŸã§ãããŸããŸãªãµã€ãºã®äºåãã¬ãŒãã³ã°æžã¿ ConvNeXt V2 ã¢ãã«ãæäŸããŠããŸããå
¬éãã¬ãŒãã³ã° ããŒã¿ã®ã¿ã䜿çšãã粟床*ã
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/convnextv2_architecture.png"
alt="æç»" width="600"/>
<small> ConvNeXt V2 ã¢ãŒããã¯ãã£ã <a href="https://arxiv.org/abs/2301.00808">å
ã®è«æ</a>ããæç²ã</small>
ãã®ã¢ãã«ã¯ [adirik](https://huggingface.co/adirik) ã«ãã£ãŠæäŸãããŸãããå
ã®ã³ãŒã㯠[ãã¡ã](https://github.com/facebookresearch/ConvNeXt-V2) ã«ãããŸãã
## Resources
ConvNeXt V2 ã®äœ¿çšãéå§ããã®ã«åœ¹ç«ã€å
¬åŒ Hugging Face ããã³ã³ãã¥ãã㣠(ð ã§ç€ºããã) ãªãœãŒã¹ã®ãªã¹ãã
<PipelineTag pipeline="image-classification"/>
- [`ConvNextV2ForImageClassification`] ã¯ããã® [ãµã³ãã« ã¹ã¯ãªãã](https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-classification) ããã³ [ããŒãããã¯](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/image_classification.ipynb)ã
ããã«å«ãããªãœãŒã¹ã®éä¿¡ã«èå³ãããå Žåã¯ããæ°è»œã«ãã« ãªã¯ãšã¹ããéããŠãã ããã審æ»ãããŠããã ããŸãããªãœãŒã¹ã¯ãæ¢åã®ãªãœãŒã¹ãè€è£œããã®ã§ã¯ãªããäœãæ°ãããã®ã瀺ãããšãçæ³çã§ãã
## ConvNextV2Config
[[autodoc]] ConvNextV2Config
## ConvNextV2Model
[[autodoc]] ConvNextV2Model
- forward
## ConvNextV2ForImageClassification
[[autodoc]] ConvNextV2ForImageClassification
- forward
## TFConvNextV2Model
[[autodoc]] TFConvNextV2Model
- call
## TFConvNextV2ForImageClassification
[[autodoc]] TFConvNextV2ForImageClassification
- call
| transformers/docs/source/ja/model_doc/convnextv2.md/0 | {
"file_path": "transformers/docs/source/ja/model_doc/convnextv2.md",
"repo_id": "transformers",
"token_count": 1916
} | 253 |
<!---
Copyright 2023 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
-->
# Model training anatomy
ã¢ãã«ãã¬ãŒãã³ã°ã®å¹çãåäžãããããã«é©çšã§ããããã©ãŒãã³ã¹æé©åãã¯ããã¯ãç解ããã«ã¯ããã¬ãŒãã³ã°äžã«GPUãã©ã®ããã«å©çšãããããããã³å®è¡ãããæäœã«å¿ããŠèšç®åŒ·åºŠãã©ã®ããã«å€åããããç解ããããšã圹ç«ã¡ãŸãã
ãŸãã¯ãGPUã®å©çšäŸãšã¢ãã«ã®ãã¬ãŒãã³ã°å®è¡ã«é¢ãã瀺åã«å¯ãäŸãæ¢æ±ããããšããå§ããŸãããããã¢ã³ã¹ãã¬ãŒã·ã§ã³ã®ããã«ãããã€ãã®ã©ã€ãã©ãªãã€ã³ã¹ããŒã«ããå¿
èŠããããŸã:
```bash
pip install transformers datasets accelerate nvidia-ml-py3
```
`nvidia-ml-py3` ã©ã€ãã©ãªã¯ãPythonå
ããã¢ãã«ã®ã¡ã¢ãªäœ¿çšç¶æ³ãã¢ãã¿ãŒããããšãå¯èœã«ããŸãããããããã¿ãŒããã«ã§ã® `nvidia-smi` ã³ãã³ãã«ã€ããŠã¯ãèããããããŸãããããã®ã©ã€ãã©ãªã䜿çšãããšãPythonããåãæ
å ±ã«ã¢ã¯ã»ã¹ã§ããŸãã
ãããããããã€ãã®ãããŒããŒã¿ãäœæããŸãã100ãã30000ã®éã®ã©ã³ãã ãªããŒã¯ã³IDãšãåé¡åšã®ããã®ãã€ããªã©ãã«ã§ããåèšã§ã512ã®ã·ãŒã±ã³ã¹ããããããããã®é·ãã¯512ã§ãPyTorchãã©ãŒãããã® [`~datasets.Dataset`] ã«æ ŒçŽãããŸãã
```py
>>> import numpy as np
>>> from datasets import Dataset
>>> seq_len, dataset_size = 512, 512
>>> dummy_data = {
... "input_ids": np.random.randint(100, 30000, (dataset_size, seq_len)),
... "labels": np.random.randint(0, 1, (dataset_size)),
... }
>>> ds = Dataset.from_dict(dummy_data)
>>> ds.set_format("pt")
```
[`Trainer`]ã䜿çšããŠGPUå©çšçãšãã¬ãŒãã³ã°å®è¡ã®èŠçŽçµ±èšæ
å ±ã衚瀺ããããã«ã2ã€ã®ãã«ããŒé¢æ°ãå®çŸ©ããŸãã
```py
>>> from pynvml import *
>>> def print_gpu_utilization():
... nvmlInit()
... handle = nvmlDeviceGetHandleByIndex(0)
... info = nvmlDeviceGetMemoryInfo(handle)
... print(f"GPU memory occupied: {info.used//1024**2} MB.")
>>> def print_summary(result):
... print(f"Time: {result.metrics['train_runtime']:.2f}")
... print(f"Samples/second: {result.metrics['train_samples_per_second']:.2f}")
... print_gpu_utilization()
```
以äžã¯ãç¡æã®GPUã¡ã¢ãªããéå§ããŠããããšã確èªããŸãããïŒ
```py
>>> print_gpu_utilization()
GPU memory occupied: 0 MB.
```
GPUã¡ã¢ãªãã¢ãã«ãèªã¿èŸŒãåã®ããã«å æãããŠããªãããã«èŠããŸãããããã䜿ãã®ãã·ã³ã§ã®ç¶æ³ã§ãªãå Žåã¯ãGPUã¡ã¢ãªã䜿çšããŠãããã¹ãŠã®ããã»ã¹ãåæ¢ããŠãã ããããã ãããã¹ãŠã®ç©ºãGPUã¡ã¢ãªããŠãŒã¶ãŒã䜿çšã§ããããã§ã¯ãããŸãããã¢ãã«ãGPUã«èªã¿èŸŒãŸãããšãã«ãŒãã«ãèªã¿èŸŒãŸãã1ã2GBã®ã¡ã¢ãªã䜿çšããããšããããŸãããããã©ãããããã確èªããããã«ãGPUã«å°ããªãã³ãœã«ãèªã¿èŸŒããšãã«ãŒãã«ãèªã¿èŸŒãŸããŸãã
```py
>>> import torch
>>> torch.ones((1, 1)).to("cuda")
>>> print_gpu_utilization()
GPU memory occupied: 1343 MB.
```
ã«ãŒãã«ã ãã§1.3GBã®GPUã¡ã¢ãªã䜿çšããŠããããšãããããŸãã次ã«ãã¢ãã«ãã©ãã ãã®ã¹ããŒã¹ã䜿çšããŠããããèŠãŠã¿ãŸãããã
## Load Model
ãŸãã`bert-large-uncased` ã¢ãã«ãèªã¿èŸŒã¿ãŸããã¢ãã«ã®éã¿ãçŽæ¥GPUã«èªã¿èŸŒãããšã§ãéã¿ã ããã©ãã ãã®ã¹ããŒã¹ã䜿çšããŠãããã確èªã§ããŸãã
```py
>>> from transformers import AutoModelForSequenceClassification
>>> model = AutoModelForSequenceClassification.from_pretrained("bert-large-uncased").to("cuda")
>>> print_gpu_utilization()
GPU memory occupied: 2631 MB.
```
ã¢ãã«ã®éã¿ã ãã§ãGPUã¡ã¢ãªã1.3 GB䜿çšããŠããããšãããããŸããæ£ç¢ºãªæ°å€ã¯ã䜿çšããŠããå
·äœçãªGPUã«äŸåããŸããæ°ããGPUã§ã¯ãã¢ãã«ã®éã¿ãæé©åãããæ¹æ³ã§èªã¿èŸŒãŸãããããã¢ãã«ã®äœ¿çšãé«éåããããšããããããã¢ãã«ãããå€ãã®ã¹ããŒã¹ãå æããããšããããŸããããŠã`nvidia-smi` CLIãšåãçµæãåŸãããããç°¡åã«ç¢ºèªããããšãã§ããŸãã
```bash
nvidia-smi
```
```bash
Tue Jan 11 08:58:05 2022
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.91.03 Driver Version: 460.91.03 CUDA Version: 11.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 Tesla V100-SXM2... On | 00000000:00:04.0 Off | 0 |
| N/A 37C P0 39W / 300W | 2631MiB / 16160MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| 0 N/A N/A 3721 C ...nvs/codeparrot/bin/python 2629MiB |
+-----------------------------------------------------------------------------+
```
ååãšåãæ°å€ãååŸãã16GBã®ã¡ã¢ãªãæèŒããV100 GPUã䜿çšããŠããããšãããããŸããããŠãã¢ãã«ã®ãã¬ãŒãã³ã°ãéå§ããGPUã¡ã¢ãªã®æ¶è²»ãã©ã®ããã«å€åãããã確èªããŠã¿ãŸãããããŸããããã€ãã®æšæºçãªãã¬ãŒãã³ã°åŒæ°ãèšå®ããŸã:
```py
default_args = {
"output_dir": "tmp",
"evaluation_strategy": "steps",
"num_train_epochs": 1,
"log_level": "error",
"report_to": "none",
}
```
<Tip>
è€æ°ã®å®éšãå®è¡ããäºå®ãããå Žåãå®éšéã§ã¡ã¢ãªãé©åã«ã¯ãªã¢ããããã«ãå®éšã®éã« Python ã«ãŒãã«ãåèµ·åããŠãã ããã
</Tip>
## Memory utilization at vanilla training
[`Trainer`] ã䜿çšããŠãGPU ããã©ãŒãã³ã¹ã®æé©åãã¯ããã¯ã䜿çšããã«ããããµã€ãº 4 ã§ã¢ãã«ããã¬ãŒãã³ã°ããŸãããïŒ
```py
>>> from transformers import TrainingArguments, Trainer, logging
>>> logging.set_verbosity_error()
>>> training_args = TrainingArguments(per_device_train_batch_size=4, **default_args)
>>> trainer = Trainer(model=model, args=training_args, train_dataset=ds)
>>> result = trainer.train()
>>> print_summary(result)
```
```
Time: 57.82
Samples/second: 8.86
GPU memory occupied: 14949 MB.
```
æ¢ã«ãæ¯èŒçå°ããããããµã€ãºã§ããGPUã®ã»ãšãã©ã®ã¡ã¢ãªããã§ã«äœ¿çšãããŠããããšãããããŸãããããããã倧ããªããããµã€ãºã䜿çšããããšã¯ããã°ãã°ã¢ãã«ã®åæãéããªã£ãããæçµçãªæ§èœãåäžãããããããšããããŸãããããã£ãŠãçæ³çã«ã¯ãããããµã€ãºãã¢ãã«ã®èŠä»¶ã«åãããŠèª¿æŽãããã®ã§ãããGPUã®å¶éã«åãããŠèª¿æŽããå¿
èŠã¯ãããŸãããèå³æ·±ãããšã«ãã¢ãã«ã®ãµã€ãºãããã¯ããã«å€ãã®ã¡ã¢ãªã䜿çšããŠããŸãããªããããªãã®ããå°ãç解ããããã«ãã¢ãã«ã®æäœãšã¡ã¢ãªã®å¿
èŠæ§ãèŠãŠã¿ãŸãããã
## Anatomy of Model's Operations
Transformerã¢ãŒããã¯ãã£ã«ã¯ãèšç®åŒ·åºŠã«ãã£ãŠä»¥äžã®3ã€ã®äž»èŠãªæäœã°ã«ãŒããå«ãŸããŠããŸãã
1. **ãã³ãœã«ã®åçž®**
ç·åœ¢å±€ãšMulti-Head Attentionã®ã³ã³ããŒãã³ãã¯ããã¹ãŠãããåŠçããã **è¡å-è¡åã®ä¹ç®** ãè¡ããŸãããããã®æäœã¯ãTransformerã®ãã¬ãŒãã³ã°ã«ãããŠæãèšç®éçŽçãªéšåã§ãã
2. **çµ±èšçæ£èŠå**
Softmaxãšå±€æ£èŠåã¯ããã³ãœã«ã®åçž®ãããèšç®è² è·ãå°ãªãã1ã€ãŸãã¯è€æ°ã® **çž®çŽæäœ** ãå«ã¿ããã®çµæãããããä»ããŠé©çšãããŸãã
3. **èŠçŽ ããšã®æŒç®å**
ãããã¯æ®ãã®æŒç®åã§ãïŒ**ãã€ã¢ã¹ãããããã¢ãŠãã掻æ§åãããã³æ®å·®æ¥ç¶** ã§ãããããã¯æãèšç®éçŽçãªæäœã§ã¯ãããŸããã
ããã©ãŒãã³ã¹ã®ããã«ããã¯ãåæããéã«ããã®ç¥èã¯åœ¹ç«ã€ããšããããŸãã
ãã®èŠçŽã¯ã[Data Movement Is All You Need: Optimizing Transformers 2020ã«é¢ããã±ãŒã¹ã¹ã¿ãã£](https://arxiv.org/abs/2007.00072)ãã掟çããŠããŸãã
## Anatomy of Model's Memory
ã¢ãã«ã®ãã¬ãŒãã³ã°ãGPUã«é
眮ãããã¢ãã«ãããã¯ããã«å€ãã®ã¡ã¢ãªã䜿çšããããšãèŠãŠããŸãããããã¯ããã¬ãŒãã³ã°äžã«GPUã¡ã¢ãªã䜿çšããå€ãã®ã³ã³ããŒãã³ããååšããããã§ããGPUã¡ã¢ãªäžã®ã³ã³ããŒãã³ãã¯ä»¥äžã®éãã§ãïŒ
1. ã¢ãã«ã®éã¿
2. ãªããã£ãã€ã¶ã®ç¶æ
3. åŸé
4. åŸé
èšç®ã®ããã«ä¿åãããååã掻æ§å
5. äžæãããã¡
6. æ©èœåºæã®ã¡ã¢ãª
éåžžãAdamWã䜿çšããŠæ··å粟床ã§ãã¬ãŒãã³ã°ãããã¢ãã«ã¯ãã¢ãã«ãã©ã¡ãŒã¿ããšã«18ãã€ããšã¢ã¯ãã£ããŒã·ã§ã³ã¡ã¢ãªãå¿
èŠã§ããæšè«ã§ã¯ãªããã£ãã€ã¶ã®ç¶æ
ãšåŸé
ã¯äžèŠã§ãã®ã§ãããããå·®ãåŒãããšãã§ããŸãããããã£ãŠãæ··å粟床ã®æšè«ã«ãããŠã¯ãã¢ãã«ãã©ã¡ãŒã¿ããšã«6ãã€ããšã¢ã¯ãã£ããŒã·ã§ã³ã¡ã¢ãªãå¿
èŠã§ãã
詳现ãèŠãŠã¿ãŸãããã
**ã¢ãã«ã®éã¿:**
- fp32ãã¬ãŒãã³ã°ã®ãã©ã¡ãŒã¿ãŒæ° * 4ãã€ã
- ããã¯ã¹ãã¬ã·ãžã§ã³ãã¬ãŒãã³ã°ã®ãã©ã¡ãŒã¿ãŒæ° * 6ãã€ãïŒã¡ã¢ãªå
ã«fp32ãšfp16ã®ã¢ãã«ãç¶æïŒ
**ãªããã£ãã€ã¶ã®ç¶æ
:**
- éåžžã®AdamWã®ãã©ã¡ãŒã¿ãŒæ° * 8ãã€ãïŒ2ã€ã®ç¶æ
ãç¶æïŒ
- 8-bit AdamWãªããã£ãã€ã¶ã®ãã©ã¡ãŒã¿ãŒæ° * 2ãã€ãïŒ[bitsandbytes](https://github.com/TimDettmers/bitsandbytes)ã®ãããªãªããã£ãã€ã¶ïŒ
- ã¢ãŒã¡ã³ã¿ã ãæã€SGDã®ãããªãªããã£ãã€ã¶ã®ãã©ã¡ãŒã¿ãŒæ° * 4ãã€ãïŒ1ã€ã®ç¶æ
ãç¶æïŒ
**åŸé
**
- fp32ãŸãã¯ããã¯ã¹ãã¬ã·ãžã§ã³ãã¬ãŒãã³ã°ã®ãã©ã¡ãŒã¿ãŒæ° * 4ãã€ãïŒåŸé
ã¯åžžã«fp32ã§ä¿æïŒ
**ãã©ã¯ãŒãã¢ã¯ãã£ããŒã·ã§ã³**
- ãµã€ãºã¯å€ãã®èŠå ã«äŸåããäž»èŠãªèŠå ã¯ã·ãŒã±ã³ã¹ã®é·ããé ãå±€ã®ãµã€ãºãããã³ããããµã€ãºã§ãã
ãã©ã¯ãŒããšããã¯ã¯ãŒãã®é¢æ°ã«ãã£ãŠæž¡ãããè¿ãããå
¥åãšåºåãããã³åŸé
èšç®ã®ããã«ä¿åããããã©ã¯ãŒãã¢ã¯ãã£ããŒã·ã§ã³ããããŸãã
**äžæçãªã¡ã¢ãª**
ããã«ãèšç®ãå®äºããåŸã«è§£æŸãããããŸããŸãªäžæå€æ°ããããŸããããããã¯äžæçã«è¿œå ã®ã¡ã¢ãªãå¿
èŠãšããOOMã«éããå¯èœæ§ããããŸãããããã£ãŠãã³ãŒãã£ã³ã°æã«ã¯ãã®ãããªäžæå€æ°ã«æŠç¥çã«èããå¿
èŠãªããªã£ããæ瀺çã«è§£æŸããããšãéåžžã«éèŠã§ãã
**æ©èœåºæã®ã¡ã¢ãª**
次ã«ããœãããŠã§ã¢ã«ã¯ç¹å¥ãªã¡ã¢ãªèŠä»¶ãããå ŽåããããŸããããšãã°ãããŒã ãµãŒãã䜿çšããŠããã¹ããçæããå ŽåããœãããŠã§ã¢ã¯è€æ°ã®å
¥åãšåºåã®ã³ããŒãç¶æããå¿
èŠããããŸãã
**`forward`ãš`backward`ã®å®è¡é床**
ç³ã¿èŸŒã¿å±€ãšç·åœ¢å±€ã§ã¯ãããã¯ã¯ãŒãã«ãã©ã¯ãŒããšæ¯ã¹ãŠ2åã®FLOPSããããäžè¬çã«ã¯çŽ2åé
ããªããŸãïŒããã¯ã¯ãŒãã®ãµã€ãºãäžäŸ¿ã§ããããšãããããããã以äžã«ãªãããšããããŸãïŒã ã¢ã¯ãã£ããŒã·ã§ã³ã¯éåžžããã³ãå¹
å¶éãããŠãããããã¯ã¯ãŒãã§ã¢ã¯ãã£ããŒã·ã§ã³ããã©ã¯ãŒããããå€ãã®ããŒã¿ãèªãããšãäžè¬çã§ãïŒããšãã°ãã¢ã¯ãã£ããŒã·ã§ã³ãã©ã¯ãŒãã¯1åèªã¿åãã1åæžã蟌ã¿ãã¢ã¯ãã£ããŒã·ã§ã³ããã¯ã¯ãŒãã¯ãã©ã¯ãŒãã®gradOutputããã³åºåã2åèªã¿åãã1åæžã蟌ã¿ãŸãïŒã
ã芧ã®éããGPUã¡ã¢ãªãç¯çŽãããæäœãé«éåã§ããå¯èœæ§ã®ããããã€ãã®å ŽæããããŸãã GPUã®å©çšãšèšç®é床ã«åœ±é¿ãäžããèŠå ãç解ããã®ã§ãããã©ãŒãã³ã¹æé©åã®æè¡ã«ã€ããŠã¯ã[åäžGPUã§ã®å¹ççãªãã¬ãŒãã³ã°ã®ããã®æ¹æ³ãšããŒã«](perf_train_gpu_one)ã®ããã¥ã¡ã³ããŒã·ã§ã³ããŒãžãåç
§ããŠãã ããã
詳现ãèŠãŠã¿ãŸãããã
| transformers/docs/source/ja/model_memory_anatomy.md/0 | {
"file_path": "transformers/docs/source/ja/model_memory_anatomy.md",
"repo_id": "transformers",
"token_count": 5980
} | 254 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.