Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
import gradio as gr
|
2 |
from transformers import AutoModel, AutoTokenizer, pipeline, AutoConfig, AutoModelForCausalLM
|
3 |
-
from huggingface_hub import
|
4 |
from transformers.modeling_utils import PreTrainedModel
|
5 |
import requests
|
6 |
import json
|
@@ -11,11 +11,11 @@ import base64
|
|
11 |
import torch
|
12 |
from torch.nn.utils import prune
|
13 |
import subprocess
|
|
|
|
|
14 |
|
15 |
-
#
|
16 |
-
|
17 |
-
models = list_models()
|
18 |
-
return models
|
19 |
|
20 |
# Ensure sentencepiece is installed
|
21 |
try:
|
@@ -23,8 +23,14 @@ try:
|
|
23 |
except ImportError:
|
24 |
subprocess.check_call(["pip", "install", "sentencepiece"])
|
25 |
|
|
|
|
|
|
|
|
|
|
|
26 |
# Function to prune a model using the "merge-kit" approach
|
27 |
-
def prune_model(llm_model_name, target_size, hf_write_token, repo_name):
|
|
|
28 |
try:
|
29 |
# Load the LLM model and tokenizer
|
30 |
llm_tokenizer = AutoTokenizer.from_pretrained(llm_model_name)
|
@@ -33,12 +39,18 @@ def prune_model(llm_model_name, target_size, hf_write_token, repo_name):
|
|
33 |
torch_dtype=torch.float16,
|
34 |
)
|
35 |
|
|
|
|
|
|
|
36 |
# Get the model config
|
37 |
config = AutoConfig.from_pretrained(llm_model_name)
|
38 |
target_num_parameters = int(config.num_parameters * (target_size / 100))
|
39 |
|
40 |
# Prune the model
|
41 |
-
pruned_model = merge_kit_prune(llm_model, target_num_parameters)
|
|
|
|
|
|
|
42 |
|
43 |
# Save the pruned model
|
44 |
api = HfApi()
|
@@ -47,6 +59,9 @@ def prune_model(llm_model_name, target_size, hf_write_token, repo_name):
|
|
47 |
pruned_model.push_to_hub(repo_id, use_auth_token=hf_write_token)
|
48 |
llm_tokenizer.push_to_hub(repo_id, use_auth_token=hf_write_token)
|
49 |
|
|
|
|
|
|
|
50 |
# Create a visualization
|
51 |
fig, ax = plt.subplots(figsize=(10, 5))
|
52 |
ax.bar(["Original", "Pruned"], [config.num_parameters, pruned_model.num_parameters])
|
@@ -57,13 +72,16 @@ def prune_model(llm_model_name, target_size, hf_write_token, repo_name):
|
|
57 |
buf.seek(0)
|
58 |
image_base64 = base64.b64encode(buf.read()).decode("utf-8")
|
59 |
|
60 |
-
return f"Pruned model saved to Hugging Face Hub in repository {repo_id}", f"data:image/png;base64,{image_base64}",
|
61 |
|
62 |
except Exception as e:
|
63 |
-
|
|
|
|
|
|
|
64 |
|
65 |
# Merge-kit Pruning Function (adjust as needed)
|
66 |
-
def merge_kit_prune(model: PreTrainedModel, target_num_parameters: int) -> PreTrainedModel:
|
67 |
"""Prunes a model using a merge-kit approach.
|
68 |
Args:
|
69 |
model (PreTrainedModel): The model to be pruned.
|
@@ -75,10 +93,11 @@ def merge_kit_prune(model: PreTrainedModel, target_num_parameters: int) -> PreTr
|
|
75 |
pruning_method = "unstructured"
|
76 |
|
77 |
# Calculate the pruning amount
|
78 |
-
|
|
|
79 |
|
80 |
# Prune the model using the selected method
|
81 |
-
for name, module in model.named_modules():
|
82 |
if isinstance(module, (torch.nn.Linear, torch.nn.Conv2d)):
|
83 |
prune.random_unstructured(module, name="weight", amount=amount)
|
84 |
|
@@ -101,8 +120,18 @@ def create_interface():
|
|
101 |
pruning_status = gr.Textbox(label="Pruning Status", interactive=False)
|
102 |
prune_button = gr.Button("Prune Model")
|
103 |
visualization = gr.Image(label="Model Size Comparison", interactive=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
|
105 |
-
prune_button.click(fn=prune_model, inputs=[llm_model_name, target_size, hf_write_token, repo_name], outputs=[pruning_status, visualization])
|
106 |
|
107 |
text_input = gr.Textbox(label="Input Text")
|
108 |
text_output = gr.Textbox(label="Generated Text")
|
@@ -124,4 +153,4 @@ def create_interface():
|
|
124 |
|
125 |
# Create and launch the Gradio interface
|
126 |
demo = create_interface()
|
127 |
-
demo.launch(share=True)
|
|
|
1 |
import gradio as gr
|
2 |
from transformers import AutoModel, AutoTokenizer, pipeline, AutoConfig, AutoModelForCausalLM
|
3 |
+
from huggingface_hub import create_repo, HfApi, list_models
|
4 |
from transformers.modeling_utils import PreTrainedModel
|
5 |
import requests
|
6 |
import json
|
|
|
11 |
import torch
|
12 |
from torch.nn.utils import prune
|
13 |
import subprocess
|
14 |
+
from tqdm import tqdm
|
15 |
+
import logging
|
16 |
|
17 |
+
# Setup logging
|
18 |
+
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
|
|
|
|
19 |
|
20 |
# Ensure sentencepiece is installed
|
21 |
try:
|
|
|
23 |
except ImportError:
|
24 |
subprocess.check_call(["pip", "install", "sentencepiece"])
|
25 |
|
26 |
+
# Function to fetch open-weight LLM models
|
27 |
+
def fetch_open_weight_models():
|
28 |
+
models = list_models()
|
29 |
+
return models
|
30 |
+
|
31 |
# Function to prune a model using the "merge-kit" approach
|
32 |
+
def prune_model(llm_model_name, target_size, hf_write_token, repo_name, progress=gr.Progress(track_tqdm=True)):
|
33 |
+
log_messages = []
|
34 |
try:
|
35 |
# Load the LLM model and tokenizer
|
36 |
llm_tokenizer = AutoTokenizer.from_pretrained(llm_model_name)
|
|
|
39 |
torch_dtype=torch.float16,
|
40 |
)
|
41 |
|
42 |
+
log_messages.append("Model and tokenizer loaded successfully.")
|
43 |
+
logging.info("Model and tokenizer loaded successfully.")
|
44 |
+
|
45 |
# Get the model config
|
46 |
config = AutoConfig.from_pretrained(llm_model_name)
|
47 |
target_num_parameters = int(config.num_parameters * (target_size / 100))
|
48 |
|
49 |
# Prune the model
|
50 |
+
pruned_model = merge_kit_prune(llm_model, target_num_parameters, progress)
|
51 |
+
|
52 |
+
log_messages.append("Model pruned successfully.")
|
53 |
+
logging.info("Model pruned successfully.")
|
54 |
|
55 |
# Save the pruned model
|
56 |
api = HfApi()
|
|
|
59 |
pruned_model.push_to_hub(repo_id, use_auth_token=hf_write_token)
|
60 |
llm_tokenizer.push_to_hub(repo_id, use_auth_token=hf_write_token)
|
61 |
|
62 |
+
log_messages.append(f"Pruned model saved to Hugging Face Hub in repository {repo_id}")
|
63 |
+
logging.info(f"Pruned model saved to Hugging Face Hub in repository {repo_id}")
|
64 |
+
|
65 |
# Create a visualization
|
66 |
fig, ax = plt.subplots(figsize=(10, 5))
|
67 |
ax.bar(["Original", "Pruned"], [config.num_parameters, pruned_model.num_parameters])
|
|
|
72 |
buf.seek(0)
|
73 |
image_base64 = base64.b64encode(buf.read()).decode("utf-8")
|
74 |
|
75 |
+
return f"Pruned model saved to Hugging Face Hub in repository {repo_id}", f"data:image/png;base64,{image_base64}", "\n".join(log_messages)
|
76 |
|
77 |
except Exception as e:
|
78 |
+
error_message = f"Error: {e}"
|
79 |
+
log_messages.append(error_message)
|
80 |
+
logging.error(error_message)
|
81 |
+
return error_message, None, "\n".join(log_messages)
|
82 |
|
83 |
# Merge-kit Pruning Function (adjust as needed)
|
84 |
+
def merge_kit_prune(model: PreTrainedModel, target_num_parameters: int, progress) -> PreTrainedModel:
|
85 |
"""Prunes a model using a merge-kit approach.
|
86 |
Args:
|
87 |
model (PreTrainedModel): The model to be pruned.
|
|
|
93 |
pruning_method = "unstructured"
|
94 |
|
95 |
# Calculate the pruning amount
|
96 |
+
total_params = sum(p.numel() for p in model.parameters())
|
97 |
+
amount = 1 - (target_num_parameters / total_params)
|
98 |
|
99 |
# Prune the model using the selected method
|
100 |
+
for name, module in tqdm(model.named_modules(), desc="Pruning", file=sys.stdout):
|
101 |
if isinstance(module, (torch.nn.Linear, torch.nn.Conv2d)):
|
102 |
prune.random_unstructured(module, name="weight", amount=amount)
|
103 |
|
|
|
120 |
pruning_status = gr.Textbox(label="Pruning Status", interactive=False)
|
121 |
prune_button = gr.Button("Prune Model")
|
122 |
visualization = gr.Image(label="Model Size Comparison", interactive=False)
|
123 |
+
progress_bar = gr.Progress()
|
124 |
+
logs_button = gr.Button("Show Logs")
|
125 |
+
logs_output = gr.Textbox(label="Logs", interactive=False)
|
126 |
+
|
127 |
+
def show_logs():
|
128 |
+
with open("pruning.log", "r") as log_file:
|
129 |
+
logs = log_file.read()
|
130 |
+
return logs
|
131 |
+
|
132 |
+
logs_button.click(fn=show_logs, outputs=logs_output)
|
133 |
|
134 |
+
prune_button.click(fn=prune_model, inputs=[llm_model_name, target_size, hf_write_token, repo_name, progress_bar], outputs=[pruning_status, visualization, logs_output])
|
135 |
|
136 |
text_input = gr.Textbox(label="Input Text")
|
137 |
text_output = gr.Textbox(label="Generated Text")
|
|
|
153 |
|
154 |
# Create and launch the Gradio interface
|
155 |
demo = create_interface()
|
156 |
+
demo.launch(share=True)
|