"""OpenAI聊天封装。"""
from __future__ import annotations
import logging
import os
import sys
from typing import (
TYPE_CHECKING,
Any,
AsyncIterator,
Callable,
Dict,
Iterator,
List,
Mapping,
Optional,
Sequence,
Tuple,
Type,
Union,
)
from langchain_core._api.deprecation import deprecated
from langchain_core.callbacks import (
AsyncCallbackManagerForLLMRun,
CallbackManagerForLLMRun,
)
from langchain_core.language_models import LanguageModelInput
from langchain_core.language_models.chat_models import (
BaseChatModel,
agenerate_from_stream,
generate_from_stream,
)
from langchain_core.language_models.llms import create_base_retry_decorator
from langchain_core.messages import (
AIMessageChunk,
BaseMessage,
BaseMessageChunk,
ChatMessageChunk,
FunctionMessageChunk,
HumanMessageChunk,
SystemMessageChunk,
ToolMessageChunk,
)
from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, ChatResult
from langchain_core.pydantic_v1 import BaseModel, Field, root_validator
from langchain_core.runnables import Runnable
from langchain_core.utils import (
get_from_dict_or_env,
get_pydantic_field_names,
)
from langchain_community.adapters.openai import (
convert_dict_to_message,
convert_message_to_dict,
)
from langchain_community.utils.openai import is_openai_v1
if TYPE_CHECKING:
import tiktoken
logger = logging.getLogger(__name__)
def _import_tiktoken() -> Any:
try:
import tiktoken
except ImportError:
raise ImportError(
"Could not import tiktoken python package. "
"This is needed in order to calculate get_token_ids. "
"Please install it with `pip install tiktoken`."
)
return tiktoken
def _create_retry_decorator(
llm: ChatOpenAI,
run_manager: Optional[
Union[AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun]
] = None,
) -> Callable[[Any], Any]:
import openai
errors = [
openai.error.Timeout,
openai.error.APIError,
openai.error.APIConnectionError,
openai.error.RateLimitError,
openai.error.ServiceUnavailableError,
]
return create_base_retry_decorator(
error_types=errors, max_retries=llm.max_retries, run_manager=run_manager
)
[docs]async def acompletion_with_retry(
llm: ChatOpenAI,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> Any:
"""使用tenacity来重试异步完成调用。"""
if is_openai_v1():
return await llm.async_client.create(**kwargs)
retry_decorator = _create_retry_decorator(llm, run_manager=run_manager)
@retry_decorator
async def _completion_with_retry(**kwargs: Any) -> Any:
# Use OpenAI's async api https://github.com/openai/openai-python#async-api
return await llm.client.acreate(**kwargs)
return await _completion_with_retry(**kwargs)
def _convert_delta_to_message_chunk(
_dict: Mapping[str, Any], default_class: Type[BaseMessageChunk]
) -> BaseMessageChunk:
role = _dict.get("role")
content = _dict.get("content") or ""
additional_kwargs: Dict = {}
if _dict.get("function_call"):
function_call = dict(_dict["function_call"])
if "name" in function_call and function_call["name"] is None:
function_call["name"] = ""
additional_kwargs["function_call"] = function_call
if _dict.get("tool_calls"):
additional_kwargs["tool_calls"] = _dict["tool_calls"]
if role == "user" or default_class == HumanMessageChunk:
return HumanMessageChunk(content=content)
elif role == "assistant" or default_class == AIMessageChunk:
return AIMessageChunk(content=content, additional_kwargs=additional_kwargs)
elif role == "system" or default_class == SystemMessageChunk:
return SystemMessageChunk(content=content)
elif role == "function" or default_class == FunctionMessageChunk:
return FunctionMessageChunk(content=content, name=_dict["name"])
elif role == "tool" or default_class == ToolMessageChunk:
return ToolMessageChunk(content=content, tool_call_id=_dict["tool_call_id"])
elif role or default_class == ChatMessageChunk:
return ChatMessageChunk(content=content, role=role) # type: ignore[arg-type]
else:
return default_class(content=content) # type: ignore[call-arg]
[docs]@deprecated(
since="0.0.10", removal="0.3.0", alternative_import="langchain_openai.ChatOpenAI"
)
class ChatOpenAI(BaseChatModel):
"""`OpenAI` 聊天大型语言模型 API。
要使用，您应该已安装 ``openai`` python 包，并且
环境变量 ``OPENAI_API_KEY`` 已设置为您的 API 密钥。
任何可以传递给 openai.create 调用的有效参数都可以传递
进来，即使在此类上没有明确保存。
示例:
.. code-block:: python
from langchain_community.chat_models import ChatOpenAI
openai = ChatOpenAI(model="gpt-3.5-turbo")
"""
@property
def lc_secrets(self) -> Dict[str, str]:
return {"openai_api_key": "OPENAI_API_KEY"}
[docs] @classmethod
def get_lc_namespace(cls) -> List[str]:
"""获取langchain对象的命名空间。"""
return ["langchain", "chat_models", "openai"]
@property
def lc_attributes(self) -> Dict[str, Any]:
attributes: Dict[str, Any] = {}
if self.openai_organization:
attributes["openai_organization"] = self.openai_organization
if self.openai_api_base:
attributes["openai_api_base"] = self.openai_api_base
if self.openai_proxy:
attributes["openai_proxy"] = self.openai_proxy
return attributes
[docs] @classmethod
def is_lc_serializable(cls) -> bool:
"""返回此模型是否可以被Langchain序列化。"""
return True
client: Any = Field(default=None, exclude=True) #: :meta private:
async_client: Any = Field(default=None, exclude=True) #: :meta private:
model_name: str = Field(default="gpt-3.5-turbo", alias="model")
"""要使用的模型名称。"""
temperature: float = 0.7
"""使用哪种采样温度。"""
model_kwargs: Dict[str, Any] = Field(default_factory=dict)
"""保存任何在`create`调用中有效但未明确指定的模型参数。"""
# When updating this to use a SecretStr
# Check for classes that derive from this class (as some of them
# may assume openai_api_key is a str)
openai_api_key: Optional[str] = Field(default=None, alias="api_key")
"""如果未提供，将自动从环境变量`OPENAI_API_KEY`中推断。"""
openai_api_base: Optional[str] = Field(default=None, alias="base_url")
"""API请求的基本URL路径，如果不使用代理或服务模拟器，请留空。"""
openai_organization: Optional[str] = Field(default=None, alias="organization")
"""如果未提供，将自环境变量`OPENAI_ORG_ID`自动推断。"""
# to support explicit proxy for OpenAI
openai_proxy: Optional[str] = None
request_timeout: Union[float, Tuple[float, float], Any, None] = Field(
default=None, alias="timeout"
)
"""请求到OpenAI完成API的超时时间。可以是浮点数、httpx.Timeout或None。"""
max_retries: int = Field(default=2)
"""生成时最大的重试次数。"""
streaming: bool = False
"""是否要流式传输结果。"""
n: int = 1
"""每个提示生成的对话完成次数。"""
max_tokens: Optional[int] = None
"""生成的令牌的最大数量。"""
tiktoken_model_name: Optional[str] = None
"""在使用这个类时，传递给tiktoken的模型名称。
Tiktoken用于计算文档中令牌的数量，以限制它们在某个特定限制之下。默认情况下，当设置为None时，这将与嵌入模型名称相同。然而，在一些情况下，您可能希望使用这个嵌入类与tiktoken不支持的模型名称一起使用。这可能包括使用Azure嵌入或使用许多提供类似OpenAI API但具有不同模型的模型提供商之一。在这些情况下，为了避免在调用tiktoken时出错，您可以在这里指定要使用的模型名称。"""
default_headers: Union[Mapping[str, str], None] = None
default_query: Union[Mapping[str, object], None] = None
# Configure a custom httpx client. See the
# [httpx documentation](https://www.python-httpx.org/api/#client) for more details.
http_client: Union[Any, None] = None
"""可选的 httpx.Client。"""
class Config:
"""此pydantic对象的配置。"""
allow_population_by_field_name = True
@root_validator(pre=True)
def build_extra(cls, values: Dict[str, Any]) -> Dict[str, Any]:
"""从传入的额外参数构建额外的kwargs。"""
all_required_field_names = get_pydantic_field_names(cls)
extra = values.get("model_kwargs", {})
for field_name in list(values):
if field_name in extra:
raise ValueError(f"Found {field_name} supplied twice.")
if field_name not in all_required_field_names:
logger.warning(
f"""WARNING! {field_name} is not default parameter.
{field_name} was transferred to model_kwargs.
Please confirm that {field_name} is what you intended."""
)
extra[field_name] = values.pop(field_name)
invalid_model_kwargs = all_required_field_names.intersection(extra.keys())
if invalid_model_kwargs:
raise ValueError(
f"Parameters {invalid_model_kwargs} should be specified explicitly. "
f"Instead they were passed in as part of `model_kwargs` parameter."
)
values["model_kwargs"] = extra
return values
@root_validator()
def validate_environment(cls, values: Dict) -> Dict:
"""验证环境中是否存在API密钥和Python包。"""
if values["n"] < 1:
raise ValueError("n must be at least 1.")
if values["n"] > 1 and values["streaming"]:
raise ValueError("n must be 1 when streaming.")
values["openai_api_key"] = get_from_dict_or_env(
values, "openai_api_key", "OPENAI_API_KEY"
)
# Check OPENAI_ORGANIZATION for backwards compatibility.
values["openai_organization"] = (
values["openai_organization"]
or os.getenv("OPENAI_ORG_ID")
or os.getenv("OPENAI_ORGANIZATION")
)
values["openai_api_base"] = values["openai_api_base"] or os.getenv(
"OPENAI_API_BASE"
)
values["openai_proxy"] = get_from_dict_or_env(
values,
"openai_proxy",
"OPENAI_PROXY",
default="",
)
try:
import openai
except ImportError:
raise ImportError(
"Could not import openai python package. "
"Please install it with `pip install openai`."
)
if is_openai_v1():
client_params = {
"api_key": values["openai_api_key"],
"organization": values["openai_organization"],
"base_url": values["openai_api_base"],
"timeout": values["request_timeout"],
"max_retries": values["max_retries"],
"default_headers": values["default_headers"],
"default_query": values["default_query"],
"http_client": values["http_client"],
}
if not values.get("client"):
values["client"] = openai.OpenAI(**client_params).chat.completions
if not values.get("async_client"):
values["async_client"] = openai.AsyncOpenAI(
**client_params
).chat.completions
elif not values.get("client"):
values["client"] = openai.ChatCompletion
else:
pass
return values
@property
def _default_params(self) -> Dict[str, Any]:
"""获取调用OpenAI API的默认参数。"""
params = {
"model": self.model_name,
"stream": self.streaming,
"n": self.n,
"temperature": self.temperature,
**self.model_kwargs,
}
if self.max_tokens is not None:
params["max_tokens"] = self.max_tokens
if self.request_timeout is not None and not is_openai_v1():
params["request_timeout"] = self.request_timeout
return params
[docs] def completion_with_retry(
self, run_manager: Optional[CallbackManagerForLLMRun] = None, **kwargs: Any
) -> Any:
"""使用tenacity来重试完成调用。"""
if is_openai_v1():
return self.client.create(**kwargs)
retry_decorator = _create_retry_decorator(self, run_manager=run_manager)
@retry_decorator
def _completion_with_retry(**kwargs: Any) -> Any:
return self.client.create(**kwargs)
return _completion_with_retry(**kwargs)
def _combine_llm_outputs(self, llm_outputs: List[Optional[dict]]) -> dict:
overall_token_usage: dict = {}
system_fingerprint = None
for output in llm_outputs:
if output is None:
# Happens in streaming
continue
token_usage = output["token_usage"]
if token_usage is not None:
for k, v in token_usage.items():
if k in overall_token_usage:
overall_token_usage[k] += v
else:
overall_token_usage[k] = v
if system_fingerprint is None:
system_fingerprint = output.get("system_fingerprint")
combined = {"token_usage": overall_token_usage, "model_name": self.model_name}
if system_fingerprint:
combined["system_fingerprint"] = system_fingerprint
return combined
def _stream(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> Iterator[ChatGenerationChunk]:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs, "stream": True}
default_chunk_class = AIMessageChunk
for chunk in self.completion_with_retry(
messages=message_dicts, run_manager=run_manager, **params
):
if not isinstance(chunk, dict):
chunk = chunk.dict()
if len(chunk["choices"]) == 0:
continue
choice = chunk["choices"][0]
chunk = _convert_delta_to_message_chunk(
choice["delta"], default_chunk_class
)
finish_reason = choice.get("finish_reason")
generation_info = (
dict(finish_reason=finish_reason) if finish_reason is not None else None
)
default_chunk_class = chunk.__class__
cg_chunk = ChatGenerationChunk(
message=chunk, generation_info=generation_info
)
if run_manager:
run_manager.on_llm_new_token(cg_chunk.text, chunk=cg_chunk)
yield cg_chunk
def _generate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[CallbackManagerForLLMRun] = None,
stream: Optional[bool] = None,
**kwargs: Any,
) -> ChatResult:
should_stream = stream if stream is not None else self.streaming
if should_stream:
stream_iter = self._stream(
messages, stop=stop, run_manager=run_manager, **kwargs
)
return generate_from_stream(stream_iter)
message_dicts, params = self._create_message_dicts(messages, stop)
params = {
**params,
**({"stream": stream} if stream is not None else {}),
**kwargs,
}
response = self.completion_with_retry(
messages=message_dicts, run_manager=run_manager, **params
)
return self._create_chat_result(response)
def _create_message_dicts(
self, messages: List[BaseMessage], stop: Optional[List[str]]
) -> Tuple[List[Dict[str, Any]], Dict[str, Any]]:
params = self._client_params
if stop is not None:
if "stop" in params:
raise ValueError("`stop` found in both the input and default params.")
params["stop"] = stop
message_dicts = [convert_message_to_dict(m) for m in messages]
return message_dicts, params
def _create_chat_result(self, response: Union[dict, BaseModel]) -> ChatResult:
generations = []
if not isinstance(response, dict):
response = response.dict()
for res in response["choices"]:
message = convert_dict_to_message(res["message"])
generation_info = dict(finish_reason=res.get("finish_reason"))
if "logprobs" in res:
generation_info["logprobs"] = res["logprobs"]
gen = ChatGeneration(
message=message,
generation_info=generation_info,
)
generations.append(gen)
token_usage = response.get("usage", {})
llm_output = {
"token_usage": token_usage,
"model_name": self.model_name,
"system_fingerprint": response.get("system_fingerprint", ""),
}
return ChatResult(generations=generations, llm_output=llm_output)
async def _astream(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> AsyncIterator[ChatGenerationChunk]:
message_dicts, params = self._create_message_dicts(messages, stop)
params = {**params, **kwargs, "stream": True}
default_chunk_class = AIMessageChunk
async for chunk in await acompletion_with_retry(
self, messages=message_dicts, run_manager=run_manager, **params
):
if not isinstance(chunk, dict):
chunk = chunk.dict()
if len(chunk["choices"]) == 0:
continue
choice = chunk["choices"][0]
chunk = _convert_delta_to_message_chunk(
choice["delta"], default_chunk_class
)
finish_reason = choice.get("finish_reason")
generation_info = (
dict(finish_reason=finish_reason) if finish_reason is not None else None
)
default_chunk_class = chunk.__class__
cg_chunk = ChatGenerationChunk(
message=chunk, generation_info=generation_info
)
if run_manager:
await run_manager.on_llm_new_token(token=cg_chunk.text, chunk=cg_chunk)
yield cg_chunk
async def _agenerate(
self,
messages: List[BaseMessage],
stop: Optional[List[str]] = None,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
stream: Optional[bool] = None,
**kwargs: Any,
) -> ChatResult:
should_stream = stream if stream is not None else self.streaming
if should_stream:
stream_iter = self._astream(
messages, stop=stop, run_manager=run_manager, **kwargs
)
return await agenerate_from_stream(stream_iter)
message_dicts, params = self._create_message_dicts(messages, stop)
params = {
**params,
**({"stream": stream} if stream is not None else {}),
**kwargs,
}
response = await acompletion_with_retry(
self, messages=message_dicts, run_manager=run_manager, **params
)
return self._create_chat_result(response)
@property
def _identifying_params(self) -> Dict[str, Any]:
"""获取识别参数。"""
return {**{"model_name": self.model_name}, **self._default_params}
@property
def _client_params(self) -> Dict[str, Any]:
"""获取用于openai客户端的参数。"""
openai_creds: Dict[str, Any] = {
"model": self.model_name,
}
if not is_openai_v1():
openai_creds.update(
{
"api_key": self.openai_api_key,
"api_base": self.openai_api_base,
"organization": self.openai_organization,
}
)
if self.openai_proxy:
import openai
openai.proxy = {"http": self.openai_proxy, "https": self.openai_proxy}
return {**self._default_params, **openai_creds}
def _get_invocation_params(
self, stop: Optional[List[str]] = None, **kwargs: Any
) -> Dict[str, Any]:
"""获取用于调用模型的参数。"""
return {
"model": self.model_name,
**super()._get_invocation_params(stop=stop),
**self._default_params,
**kwargs,
}
@property
def _llm_type(self) -> str:
"""聊天模型的返回类型。"""
return "openai-chat"
def _get_encoding_model(self) -> Tuple[str, tiktoken.Encoding]:
tiktoken_ = _import_tiktoken()
if self.tiktoken_model_name is not None:
model = self.tiktoken_model_name
else:
model = self.model_name
if model == "gpt-3.5-turbo":
# gpt-3.5-turbo may change over time.
# Returning num tokens assuming gpt-3.5-turbo-0301.
model = "gpt-3.5-turbo-0301"
elif model == "gpt-4":
# gpt-4 may change over time.
# Returning num tokens assuming gpt-4-0314.
model = "gpt-4-0314"
# Returns the number of tokens used by a list of messages.
try:
encoding = tiktoken_.encoding_for_model(model)
except KeyError:
logger.warning("Warning: model not found. Using cl100k_base encoding.")
model = "cl100k_base"
encoding = tiktoken_.get_encoding(model)
return model, encoding
[docs] def get_token_ids(self, text: str) -> List[int]:
"""使用tiktoken包获取文本中的标记。"""
# tiktoken NOT supported for Python 3.7 or below
if sys.version_info[1] <= 7:
return super().get_token_ids(text)
_, encoding_model = self._get_encoding_model()
return encoding_model.encode(text)
[docs] def get_num_tokens_from_messages(self, messages: List[BaseMessage]) -> int:
"""使用tiktoken包计算gpt-3.5-turbo和gpt-4的num tokens。
官方文档：https://github.com/openai/openai-cookbook/blob/
main/examples/How_to_format_inputs_to_ChatGPT_models.ipynb
"""
if sys.version_info[1] <= 7:
return super().get_num_tokens_from_messages(messages)
model, encoding = self._get_encoding_model()
if model.startswith("gpt-3.5-turbo-0301"):
# every message follows <im_start>{role/name}\n{content}<im_end>\n
tokens_per_message = 4
# if there's a name, the role is omitted
tokens_per_name = -1
elif model.startswith("gpt-3.5-turbo") or model.startswith("gpt-4"):
tokens_per_message = 3
tokens_per_name = 1
else:
raise NotImplementedError(
f"get_num_tokens_from_messages() is not presently implemented "
f"for model {model}."
"See https://github.com/openai/openai-python/blob/main/chatml.md for "
"information on how messages are converted to tokens."
)
num_tokens = 0
messages_dict = [convert_message_to_dict(m) for m in messages]
for message in messages_dict:
num_tokens += tokens_per_message
for key, value in message.items():
# Cast str(value) in case the message value is not a string
# This occurs with function messages
num_tokens += len(encoding.encode(str(value)))
if key == "name":
num_tokens += tokens_per_name
# every reply is primed with <im_start>assistant
num_tokens += 3
return num_tokens
[docs] def bind_functions(
self,
functions: Sequence[Union[Dict[str, Any], Type[BaseModel], Callable]],
function_call: Optional[str] = None,
**kwargs: Any,
) -> Runnable[LanguageModelInput, BaseMessage]:
"""将函数（和其他对象）绑定到此聊天模型。
参数：
functions：要绑定到此聊天模型的函数定义列表。
可以是字典、pydantic模型或可调用对象。Pydantic
模型和可调用对象将自动转换为
它们的模式字典表示。
function_call：要求模型调用的函数。
必须是提供的单个函数的名称或
"auto"，以自动确定要调用的函数
（如果有的话）。
kwargs：要传递给
:class:`~langchain.runnable.Runnable` 构造函数的任何额外参数。
"""
from langchain.chains.openai_functions.base import convert_to_openai_function
formatted_functions = [convert_to_openai_function(fn) for fn in functions]
if function_call is not None:
if len(formatted_functions) != 1:
raise ValueError(
"When specifying `function_call`, you must provide exactly one "
"function."
)
if formatted_functions[0]["name"] != function_call:
raise ValueError(
f"Function call {function_call} was specified, but the only "
f"provided function was {formatted_functions[0]['name']}."
)
function_call_ = {"name": function_call}
kwargs = {**kwargs, "function_call": function_call_}
return super().bind(
functions=formatted_functions,
**kwargs,
)