Spaces:
Runtime error
Runtime error
File size: 14,199 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 |
import asyncio
import logging
from concurrent.futures import ThreadPoolExecutor
from typing import Any, Dict, List, Mapping, Optional, Sequence, TypedDict
import aiohttp
import requests
from langchain_core.callbacks import (
AsyncCallbackManagerForLLMRun,
CallbackManagerForLLMRun,
)
from langchain_core.language_models.llms import BaseLLM
from langchain_core.outputs import Generation, LLMResult
from langchain_core.pydantic_v1 import Extra, Field, root_validator
from langchain_core.utils import get_from_dict_or_env
from langchain_community.llms.utils import enforce_stop_tokens
class TrainResult(TypedDict):
"""Train result."""
loss: float
class GradientLLM(BaseLLM):
"""Gradient.ai LLM Endpoints.
GradientLLM is a class to interact with LLMs on gradient.ai
To use, set the environment variable ``GRADIENT_ACCESS_TOKEN`` with your
API token and ``GRADIENT_WORKSPACE_ID`` for your gradient workspace,
or alternatively provide them as keywords to the constructor of this class.
Example:
.. code-block:: python
from langchain_community.llms import GradientLLM
GradientLLM(
model="99148c6d-c2a0-4fbe-a4a7-e7c05bdb8a09_base_ml_model",
model_kwargs={
"max_generated_token_count": 128,
"temperature": 0.75,
"top_p": 0.95,
"top_k": 20,
"stop": [],
},
gradient_workspace_id="12345614fc0_workspace",
gradient_access_token="gradientai-access_token",
)
"""
model_id: str = Field(alias="model", min_length=2)
"Underlying gradient.ai model id (base or fine-tuned)."
gradient_workspace_id: Optional[str] = None
"Underlying gradient.ai workspace_id."
gradient_access_token: Optional[str] = None
"""gradient.ai API Token, which can be generated by going to
https://auth.gradient.ai/select-workspace
and selecting "Access tokens" under the profile drop-down.
"""
model_kwargs: Optional[dict] = None
"""Keyword arguments to pass to the model."""
gradient_api_url: str = "https://api.gradient.ai/api"
"""Endpoint URL to use."""
aiosession: Optional[aiohttp.ClientSession] = None #: :meta private:
"""ClientSession, private, subject to change in upcoming releases."""
# LLM call kwargs
class Config:
"""Configuration for this pydantic object."""
allow_population_by_field_name = True
extra = Extra.forbid
@root_validator(allow_reuse=True)
def validate_environment(cls, values: Dict) -> Dict:
"""Validate that api key and python package exists in environment."""
values["gradient_access_token"] = get_from_dict_or_env(
values, "gradient_access_token", "GRADIENT_ACCESS_TOKEN"
)
values["gradient_workspace_id"] = get_from_dict_or_env(
values, "gradient_workspace_id", "GRADIENT_WORKSPACE_ID"
)
if (
values["gradient_access_token"] is None
or len(values["gradient_access_token"]) < 10
):
raise ValueError("env variable `GRADIENT_ACCESS_TOKEN` must be set")
if (
values["gradient_workspace_id"] is None
or len(values["gradient_access_token"]) < 3
):
raise ValueError("env variable `GRADIENT_WORKSPACE_ID` must be set")
if values["model_kwargs"]:
kw = values["model_kwargs"]
if not 0 <= kw.get("temperature", 0.5) <= 1:
raise ValueError("`temperature` must be in the range [0.0, 1.0]")
if not 0 <= kw.get("top_p", 0.5) <= 1:
raise ValueError("`top_p` must be in the range [0.0, 1.0]")
if 0 >= kw.get("top_k", 0.5):
raise ValueError("`top_k` must be positive")
if 0 >= kw.get("max_generated_token_count", 1):
raise ValueError("`max_generated_token_count` must be positive")
values["gradient_api_url"] = get_from_dict_or_env(
values, "gradient_api_url", "GRADIENT_API_URL"
)
try:
import gradientai # noqa
except ImportError:
logging.warning(
"DeprecationWarning: `GradientLLM` will use "
"`pip install gradientai` in future releases of langchain."
)
except Exception:
pass
return values
@property
def _identifying_params(self) -> Mapping[str, Any]:
"""Get the identifying parameters."""
_model_kwargs = self.model_kwargs or {}
return {
**{"gradient_api_url": self.gradient_api_url},
**{"model_kwargs": _model_kwargs},
}
@property
def _llm_type(self) -> str:
"""Return type of llm."""
return "gradient"
def _kwargs_post_fine_tune_request(
self, inputs: Sequence[str], kwargs: Mapping[str, Any]
) -> Mapping[str, Any]:
"""Build the kwargs for the Post request, used by sync
Args:
prompt (str): prompt used in query
kwargs (dict): model kwargs in payload
Returns:
Dict[str, Union[str,dict]]: _description_
"""
_model_kwargs = self.model_kwargs or {}
_params = {**_model_kwargs, **kwargs}
multipliers = _params.get("multipliers", None)
return dict(
url=f"{self.gradient_api_url}/models/{self.model_id}/fine-tune",
headers={
"authorization": f"Bearer {self.gradient_access_token}",
"x-gradient-workspace-id": f"{self.gradient_workspace_id}",
"accept": "application/json",
"content-type": "application/json",
},
json=dict(
samples=tuple(
{
"inputs": input,
}
for input in inputs
)
if multipliers is None
else tuple(
{
"inputs": input,
"fineTuningParameters": {
"multiplier": multiplier,
},
}
for input, multiplier in zip(inputs, multipliers)
),
),
)
def _kwargs_post_request(
self, prompt: str, kwargs: Mapping[str, Any]
) -> Mapping[str, Any]:
"""Build the kwargs for the Post request, used by sync
Args:
prompt (str): prompt used in query
kwargs (dict): model kwargs in payload
Returns:
Dict[str, Union[str,dict]]: _description_
"""
_model_kwargs = self.model_kwargs or {}
_params = {**_model_kwargs, **kwargs}
return dict(
url=f"{self.gradient_api_url}/models/{self.model_id}/complete",
headers={
"authorization": f"Bearer {self.gradient_access_token}",
"x-gradient-workspace-id": f"{self.gradient_workspace_id}",
"accept": "application/json",
"content-type": "application/json",
},
json=dict(
query=prompt,
maxGeneratedTokenCount=_params.get("max_generated_token_count", None),
temperature=_params.get("temperature", None),
topK=_params.get("top_k", None),
topP=_params.get("top_p", None),
),
)
def _call(
self,
prompt: str,
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> str:
"""Call to Gradients API `model/{id}/complete`.
Args:
prompt: The prompt to pass into the model.
stop: Optional list of stop words to use when generating.
Returns:
The string generated by the model.
"""
try:
response = requests.post(**self._kwargs_post_request(prompt, kwargs))
if response.status_code != 200:
raise Exception(
f"Gradient returned an unexpected response with status "
f"{response.status_code}: {response.text}"
)
except requests.exceptions.RequestException as e:
raise Exception(f"RequestException while calling Gradient Endpoint: {e}")
text = response.json()["generatedOutput"]
if stop is not None:
# Apply stop tokens when making calls to Gradient
text = enforce_stop_tokens(text, stop)
return text
async def _acall(
self,
prompt: str,
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> str:
"""Async Call to Gradients API `model/{id}/complete`.
Args:
prompt: The prompt to pass into the model.
stop: Optional list of stop words to use when generating.
Returns:
The string generated by the model.
"""
if not self.aiosession:
async with aiohttp.ClientSession() as session:
async with session.post(
**self._kwargs_post_request(prompt=prompt, kwargs=kwargs)
) as response:
if response.status != 200:
raise Exception(
f"Gradient returned an unexpected response with status "
f"{response.status}: {response.text}"
)
text = (await response.json())["generatedOutput"]
else:
async with self.aiosession.post(
**self._kwargs_post_request(prompt=prompt, kwargs=kwargs)
) as response:
if response.status != 200:
raise Exception(
f"Gradient returned an unexpected response with status "
f"{response.status}: {response.text}"
)
text = (await response.json())["generatedOutput"]
if stop is not None:
# Apply stop tokens when making calls to Gradient
text = enforce_stop_tokens(text, stop)
return text
def _generate(
self,
prompts: List[str],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> LLMResult:
"""Run the LLM on the given prompt and input."""
# same thing with threading
def _inner_generate(prompt: str) -> List[Generation]:
return [
Generation(
text=self._call(
prompt=prompt, stop=stop, run_manager=run_manager, **kwargs
)
)
]
if len(prompts) <= 1:
generations = list(map(_inner_generate, prompts))
else:
with ThreadPoolExecutor(min(8, len(prompts))) as p:
generations = list(p.map(_inner_generate, prompts))
return LLMResult(generations=generations)
async def _agenerate(
self,
prompts: List[str],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> LLMResult:
"""Run the LLM on the given prompt and input."""
generations = []
for generation in asyncio.gather(
[self._acall(prompt, stop=stop, run_manager=run_manager, **kwargs)]
for prompt in prompts
):
generations.append([Generation(text=generation)])
return LLMResult(generations=generations)
def train_unsupervised(
self,
inputs: Sequence[str],
**kwargs: Any,
) -> TrainResult:
try:
response = requests.post(
**self._kwargs_post_fine_tune_request(inputs, kwargs)
)
if response.status_code != 200:
raise Exception(
f"Gradient returned an unexpected response with status "
f"{response.status_code}: {response.text}"
)
except requests.exceptions.RequestException as e:
raise Exception(f"RequestException while calling Gradient Endpoint: {e}")
response_json = response.json()
loss = response_json["sumLoss"] / response_json["numberOfTrainableTokens"]
return TrainResult(loss=loss)
async def atrain_unsupervised(
self,
inputs: Sequence[str],
**kwargs: Any,
) -> TrainResult:
if not self.aiosession:
async with aiohttp.ClientSession() as session:
async with session.post(
**self._kwargs_post_fine_tune_request(inputs, kwargs)
) as response:
if response.status != 200:
raise Exception(
f"Gradient returned an unexpected response with status "
f"{response.status}: {response.text}"
)
response_json = await response.json()
loss = (
response_json["sumLoss"]
/ response_json["numberOfTrainableTokens"]
)
else:
async with self.aiosession.post(
**self._kwargs_post_fine_tune_request(inputs, kwargs)
) as response:
if response.status != 200:
raise Exception(
f"Gradient returned an unexpected response with status "
f"{response.status}: {response.text}"
)
response_json = await response.json()
loss = (
response_json["sumLoss"] / response_json["numberOfTrainableTokens"]
)
return TrainResult(loss=loss)
|