import logging
from typing import Any, Dict, List, Optional
from langchain_core.callbacks import CallbackManagerForLLMRun
from langchain_core.language_models.llms import LLM
from langchain_core.outputs import Generation, LLMResult
from langchain_core.pydantic_v1 import Extra, Field, root_validator
from langchain_community.llms.utils import enforce_stop_tokens
logger = logging.getLogger(__name__)
EXAMPLE_URL = "https://clarifai.com/openai/chat-completion/models/GPT-4"
[docs]class Clarifai(LLM):
"""Clarifai大型语言模型。
要使用,您应该在Clarifai平台上拥有一个帐户,
安装``clarifai`` python包,
并设置环境变量``CLARIFAI_PAT``为您的PAT密钥,
或将其作为命名参数传递给构造函数。
示例:
.. code-block:: python
from langchain_community.llms import Clarifai
clarifai_llm = Clarifai(user_id=USER_ID, app_id=APP_ID, model_id=MODEL_ID)
(或)
clarifai_llm = Clarifai(model_url=EXAMPLE_URL)
"""
model_url: Optional[str] = None
"""要使用的模型URL。"""
model_id: Optional[str] = None
"""要使用的模型ID。"""
model_version_id: Optional[str] = None
"""要使用的模型版本ID。"""
app_id: Optional[str] = None
"""用于Clarifai应用程序的应用程序ID。"""
user_id: Optional[str] = None
"""用于Clarifai的用户ID。"""
pat: Optional[str] = Field(default=None, exclude=True) #: :meta private:
"""用于使用Clarifai的个人访问令牌。"""
token: Optional[str] = Field(default=None, exclude=True) #: :meta private:
"""用于使用Clarifai会话令牌。"""
model: Any = Field(default=None, exclude=True) #: :meta private:
api_base: str = "https://api.clarifai.com"
class Config:
"""此pydantic对象的配置。"""
extra = Extra.forbid
@root_validator()
def validate_environment(cls, values: Dict) -> Dict:
"""验证我们是否具有访问Clarifai平台所需的所有必要信息,并且Python包存在于环境中。
"""
try:
from clarifai.client.model import Model
except ImportError:
raise ImportError(
"Could not import clarifai python package. "
"Please install it with `pip install clarifai`."
)
user_id = values.get("user_id")
app_id = values.get("app_id")
model_id = values.get("model_id")
model_version_id = values.get("model_version_id")
model_url = values.get("model_url")
api_base = values.get("api_base")
pat = values.get("pat")
token = values.get("token")
values["model"] = Model(
url=model_url,
app_id=app_id,
user_id=user_id,
model_version=dict(id=model_version_id),
pat=pat,
token=token,
model_id=model_id,
base_url=api_base,
)
return values
@property
def _default_params(self) -> Dict[str, Any]:
"""获取调用Clarifai API的默认参数。"""
return {}
@property
def _identifying_params(self) -> Dict[str, Any]:
"""获取识别参数。"""
return {
**{
"model_url": self.model_url,
"user_id": self.user_id,
"app_id": self.app_id,
"model_id": self.model_id,
}
}
@property
def _llm_type(self) -> str:
"""llm的返回类型。"""
return "clarifai"
def _call(
self,
prompt: str,
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
inference_params: Optional[Dict[str, Any]] = None,
**kwargs: Any,
) -> str:
"""调用Clarifai的PostModelOutputs端点。
参数:
prompt: 传递给模型的提示。
stop: 生成时可选的停用词列表。
返回:
模型生成的字符串。
示例:
.. code-block:: python
response = clarifai_llm.invoke("Tell me a joke.")
"""
try:
(inference_params := {}) if inference_params is None else inference_params
predict_response = self.model.predict_by_bytes(
bytes(prompt, "utf-8"),
input_type="text",
inference_params=inference_params,
)
text = predict_response.outputs[0].data.text.raw
if stop is not None:
text = enforce_stop_tokens(text, stop)
except Exception as e:
logger.error(f"Predict failed, exception: {e}")
return text
def _generate(
self,
prompts: List[str],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
inference_params: Optional[Dict[str, Any]] = None,
**kwargs: Any,
) -> LLMResult:
"""在给定的提示和输入上运行LLM。"""
# TODO: add caching here.
try:
from clarifai.client.input import Inputs
except ImportError:
raise ImportError(
"Could not import clarifai python package. "
"Please install it with `pip install clarifai`."
)
generations = []
batch_size = 32
input_obj = Inputs.from_auth_helper(self.model.auth_helper)
try:
for i in range(0, len(prompts), batch_size):
batch = prompts[i : i + batch_size]
input_batch = [
input_obj.get_text_input(input_id=str(id), raw_text=inp)
for id, inp in enumerate(batch)
]
(
inference_params := {}
) if inference_params is None else inference_params
predict_response = self.model.predict(
inputs=input_batch, inference_params=inference_params
)
for output in predict_response.outputs:
if stop is not None:
text = enforce_stop_tokens(output.data.text.raw, stop)
else:
text = output.data.text.raw
generations.append([Generation(text=text)])
except Exception as e:
logger.error(f"Predict failed, exception: {e}")
return LLMResult(generations=generations)