Spaces:
Runtime error
Runtime error
File size: 9,521 Bytes
ed4d993 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 |
import logging
from typing import Any, List, Mapping, Optional
from langchain_core.callbacks import CallbackManagerForLLMRun
from langchain_core.language_models.llms import LLM
from langchain_core.pydantic_v1 import Extra
DEFAULT_MODEL_ID = "gpt2"
logger = logging.getLogger(__name__)
class IpexLLM(LLM):
"""IpexLLM model.
Example:
.. code-block:: python
from langchain_community.llms import IpexLLM
llm = IpexLLM.from_model_id(model_id="THUDM/chatglm-6b")
"""
model_id: str = DEFAULT_MODEL_ID
"""Model name or model path to use."""
model_kwargs: Optional[dict] = None
"""Keyword arguments passed to the model."""
model: Any #: :meta private:
"""IpexLLM model."""
tokenizer: Any #: :meta private:
"""Huggingface tokenizer model."""
streaming: bool = True
"""Whether to stream the results, token by token."""
class Config:
"""Configuration for this pydantic object."""
extra = Extra.forbid
@classmethod
def from_model_id(
cls,
model_id: str,
model_kwargs: Optional[dict] = None,
*,
tokenizer_id: Optional[str] = None,
load_in_4bit: bool = True,
load_in_low_bit: Optional[str] = None,
**kwargs: Any,
) -> LLM:
"""
Construct object from model_id
Args:
model_id: Path for the huggingface repo id to be downloaded or
the huggingface checkpoint folder.
tokenizer_id: Path for the huggingface repo id to be downloaded or
the huggingface checkpoint folder which contains the tokenizer.
load_in_4bit: "Whether to load model in 4bit.
Unused if `load_in_low_bit` is not None.
load_in_low_bit: Which low bit precisions to use when loading model.
Example values: 'sym_int4', 'asym_int4', 'fp4', 'nf4', 'fp8', etc.
Overrides `load_in_4bit` if specified.
model_kwargs: Keyword arguments to pass to the model and tokenizer.
kwargs: Extra arguments to pass to the model and tokenizer.
Returns:
An object of IpexLLM.
"""
return cls._load_model(
model_id=model_id,
tokenizer_id=tokenizer_id,
low_bit_model=False,
load_in_4bit=load_in_4bit,
load_in_low_bit=load_in_low_bit,
model_kwargs=model_kwargs,
kwargs=kwargs,
)
@classmethod
def from_model_id_low_bit(
cls,
model_id: str,
model_kwargs: Optional[dict] = None,
*,
tokenizer_id: Optional[str] = None,
**kwargs: Any,
) -> LLM:
"""
Construct low_bit object from model_id
Args:
model_id: Path for the ipex-llm transformers low-bit model folder.
tokenizer_id: Path for the huggingface repo id or local model folder
which contains the tokenizer.
model_kwargs: Keyword arguments to pass to the model and tokenizer.
kwargs: Extra arguments to pass to the model and tokenizer.
Returns:
An object of IpexLLM.
"""
return cls._load_model(
model_id=model_id,
tokenizer_id=tokenizer_id,
low_bit_model=True,
load_in_4bit=False, # not used for low-bit model
load_in_low_bit=None, # not used for low-bit model
model_kwargs=model_kwargs,
kwargs=kwargs,
)
@classmethod
def _load_model(
cls,
model_id: str,
tokenizer_id: Optional[str] = None,
load_in_4bit: bool = False,
load_in_low_bit: Optional[str] = None,
low_bit_model: bool = False,
model_kwargs: Optional[dict] = None,
kwargs: Optional[dict] = None,
) -> Any:
try:
from ipex_llm.transformers import (
AutoModel,
AutoModelForCausalLM,
)
from transformers import AutoTokenizer, LlamaTokenizer
except ImportError:
raise ImportError(
"Could not import ipex-llm. "
"Please install `ipex-llm` properly following installation guides: "
"https://github.com/intel-analytics/ipex-llm?tab=readme-ov-file#install-ipex-llm."
)
_model_kwargs = model_kwargs or {}
kwargs = kwargs or {}
_tokenizer_id = tokenizer_id or model_id
try:
tokenizer = AutoTokenizer.from_pretrained(_tokenizer_id, **_model_kwargs)
except Exception:
tokenizer = LlamaTokenizer.from_pretrained(_tokenizer_id, **_model_kwargs)
# restore model_kwargs
if "trust_remote_code" in _model_kwargs:
_model_kwargs = {
k: v for k, v in _model_kwargs.items() if k != "trust_remote_code"
}
# load model with AutoModelForCausalLM and falls back to AutoModel on failure.
load_kwargs = {
"use_cache": True,
"trust_remote_code": True,
}
if not low_bit_model:
if load_in_low_bit is not None:
load_function_name = "from_pretrained"
load_kwargs["load_in_low_bit"] = load_in_low_bit # type: ignore
else:
load_function_name = "from_pretrained"
load_kwargs["load_in_4bit"] = load_in_4bit
else:
load_function_name = "load_low_bit"
try:
# Attempt to load with AutoModelForCausalLM
model = cls._load_model_general(
AutoModelForCausalLM,
load_function_name=load_function_name,
model_id=model_id,
load_kwargs=load_kwargs,
model_kwargs=_model_kwargs,
)
except Exception:
# Fallback to AutoModel if there's an exception
model = cls._load_model_general(
AutoModel,
load_function_name=load_function_name,
model_id=model_id,
load_kwargs=load_kwargs,
model_kwargs=_model_kwargs,
)
return cls(
model_id=model_id,
model=model,
tokenizer=tokenizer,
model_kwargs=_model_kwargs,
**kwargs,
)
@staticmethod
def _load_model_general(
model_class: Any,
load_function_name: str,
model_id: str,
load_kwargs: dict,
model_kwargs: dict,
) -> Any:
"""General function to attempt to load a model."""
try:
load_function = getattr(model_class, load_function_name)
return load_function(model_id, **{**load_kwargs, **model_kwargs})
except Exception as e:
logger.error(
f"Failed to load model using "
f"{model_class.__name__}.{load_function_name}: {e}"
)
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
return {
"model_id": self.model_id,
"model_kwargs": self.model_kwargs,
}
@property
def _llm_type(self) -> str:
return "ipex-llm"
def _call(
self,
prompt: str,
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> str:
if self.streaming:
from transformers import TextStreamer
input_ids = self.tokenizer.encode(prompt, return_tensors="pt")
streamer = TextStreamer(
self.tokenizer, skip_prompt=True, skip_special_tokens=True
)
if stop is not None:
from transformers.generation.stopping_criteria import (
StoppingCriteriaList,
)
from transformers.tools.agents import StopSequenceCriteria
# stop generation when stop words are encountered
# TODO: stop generation when the following one is stop word
stopping_criteria = StoppingCriteriaList(
[StopSequenceCriteria(stop, self.tokenizer)]
)
else:
stopping_criteria = None
output = self.model.generate(
input_ids,
streamer=streamer,
stopping_criteria=stopping_criteria,
**kwargs,
)
text = self.tokenizer.decode(output[0], skip_special_tokens=True)
return text
else:
input_ids = self.tokenizer.encode(prompt, return_tensors="pt")
if stop is not None:
from transformers.generation.stopping_criteria import (
StoppingCriteriaList,
)
from transformers.tools.agents import StopSequenceCriteria
stopping_criteria = StoppingCriteriaList(
[StopSequenceCriteria(stop, self.tokenizer)]
)
else:
stopping_criteria = None
output = self.model.generate(
input_ids, stopping_criteria=stopping_criteria, **kwargs
)
text = self.tokenizer.decode(output[0], skip_special_tokens=True)[
len(prompt) :
]
return text
|