"""deepinfra.com chat models wrapper"""
from __future__ import annotations
import json
import logging
from json import JSONDecodeError
from typing import (
Any,
AsyncIterator,
Callable,
Dict,
Iterator,
List,
Mapping,
Optional,
Sequence,
Tuple,
Type,
Union,
)
import aiohttp
import requests
from langchain_core.callbacks.manager 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 (
AIMessage,
AIMessageChunk,
BaseMessage,
BaseMessageChunk,
ChatMessage,
ChatMessageChunk,
FunctionMessage,
FunctionMessageChunk,
HumanMessage,
HumanMessageChunk,
SystemMessage,
SystemMessageChunk,
ToolMessage,
)
from langchain_core.messages.tool import ToolCall
from langchain_core.messages.tool import tool_call as create_tool_call
from langchain_core.outputs import (
ChatGeneration,
ChatGenerationChunk,
ChatResult,
)
from langchain_core.runnables import Runnable
from langchain_core.tools import BaseTool
from langchain_core.utils import get_from_dict_or_env
from langchain_core.utils.function_calling import convert_to_openai_tool
from pydantic import BaseModel, ConfigDict, Field, model_validator
from typing_extensions import Self
from langchain_community.utilities.requests import Requests
logger = logging.getLogger(__name__)
[docs]
class ChatDeepInfraException(Exception):
"""Exception raised when the DeepInfra API returns an error."""
pass
def _create_retry_decorator(
llm: ChatDeepInfra,
run_manager: Optional[
Union[AsyncCallbackManagerForLLMRun, CallbackManagerForLLMRun]
] = None,
) -> Callable[[Any], Any]:
"""Returns a tenacity retry decorator, preconfigured to handle PaLM exceptions."""
return create_base_retry_decorator(
error_types=[requests.exceptions.ConnectTimeout, ChatDeepInfraException],
max_retries=llm.max_retries,
run_manager=run_manager,
)
def _parse_tool_calling(tool_call: dict) -> ToolCall:
"""
Convert a tool calling response from server to a ToolCall object.
Args:
tool_call:
Returns:
"""
name = tool_call["function"].get("name", "")
try:
args = json.loads(tool_call["function"]["arguments"])
except (JSONDecodeError, TypeError):
args = {}
id = tool_call.get("id")
return create_tool_call(name=name, args=args, id=id)
def _convert_to_tool_calling(tool_call: ToolCall) -> Dict[str, Any]:
"""
Convert a ToolCall object to a tool calling request for server.
Args:
tool_call:
Returns:
"""
return {
"type": "function",
"function": {
"arguments": json.dumps(tool_call["args"]),
"name": tool_call["name"],
},
"id": tool_call.get("id"),
}
def _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
role = _dict["role"]
if role == "user":
return HumanMessage(content=_dict["content"])
elif role == "assistant":
content = _dict.get("content", "") or ""
tool_calls_content = _dict.get("tool_calls", []) or []
tool_calls = [
_parse_tool_calling(tool_call) for tool_call in tool_calls_content
]
return AIMessage(content=content, tool_calls=tool_calls)
elif role == "system":
return SystemMessage(content=_dict["content"])
elif role == "function":
return FunctionMessage(content=_dict["content"], name=_dict["name"])
else:
return ChatMessage(content=_dict["content"], role=role)
def _convert_delta_to_message_chunk(
_dict: Mapping[str, Any], default_class: Type[BaseMessageChunk]
) -> BaseMessageChunk:
role = _dict.get("role")
content = _dict.get("content") or ""
tool_calls = _dict.get("tool_calls") or []
if role == "user" or default_class == HumanMessageChunk:
return HumanMessageChunk(content=content)
elif role == "assistant" or default_class == AIMessageChunk:
tool_calls = [_parse_tool_calling(tool_call) for tool_call in tool_calls]
return AIMessageChunk(content=content, tool_calls=tool_calls)
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 or default_class == ChatMessageChunk:
return ChatMessageChunk(content=content, role=role) # type: ignore[arg-type]
else:
return default_class(content=content) # type: ignore[call-arg]
def _convert_message_to_dict(message: BaseMessage) -> dict:
if isinstance(message, ChatMessage):
message_dict = {"role": message.role, "content": message.content}
elif isinstance(message, HumanMessage):
message_dict = {"role": "user", "content": message.content}
elif isinstance(message, AIMessage):
tool_calls = [
_convert_to_tool_calling(tool_call) for tool_call in message.tool_calls
]
message_dict = {
"role": "assistant",
"content": message.content,
"tool_calls": tool_calls, # type: ignore[dict-item]
}
elif isinstance(message, SystemMessage):
message_dict = {"role": "system", "content": message.content}
elif isinstance(message, FunctionMessage):
message_dict = {
"role": "function",
"content": message.content,
"name": message.name,
}
elif isinstance(message, ToolMessage):
message_dict = {
"role": "tool",
"content": message.content,
"name": message.name, # type: ignore[dict-item]
"tool_call_id": message.tool_call_id,
}
else:
raise ValueError(f"Got unknown type {message}")
if "name" in message.additional_kwargs:
message_dict["name"] = message.additional_kwargs["name"]
return message_dict
[docs]
class ChatDeepInfra(BaseChatModel):
"""A chat model that uses the DeepInfra API."""
# client: Any #: :meta private:
model_name: str = Field(default="meta-llama/Llama-2-70b-chat-hf", alias="model")
"""Model name to use."""
url: str = "https://api.deepinfra.com/v1/openai/chat/completions"
"""URL to use for the API call."""
deepinfra_api_token: Optional[str] = None
request_timeout: Optional[float] = Field(default=None, alias="timeout")
temperature: Optional[float] = 1
model_kwargs: Dict[str, Any] = Field(default_factory=dict)
"""Run inference with this temperature. Must be in the closed
interval [0.0, 1.0]."""
top_p: Optional[float] = None
"""Decode using nucleus sampling: consider the smallest set of tokens whose
probability sum is at least top_p. Must be in the closed interval [0.0, 1.0]."""
top_k: Optional[int] = None
"""Decode using top-k sampling: consider the set of top_k most probable tokens.
Must be positive."""
n: int = 1
"""Number of chat completions to generate for each prompt. Note that the API may
not return the full n completions if duplicates are generated."""
max_tokens: int = 256
streaming: bool = False
max_retries: int = 1
model_config = ConfigDict(
populate_by_name=True,
)
@property
def _default_params(self) -> Dict[str, Any]:
"""Get the default parameters for calling OpenAI API."""
return {
"model": self.model_name,
"max_tokens": self.max_tokens,
"stream": self.streaming,
"n": self.n,
"temperature": self.temperature,
"request_timeout": self.request_timeout,
**self.model_kwargs,
}
@property
def _client_params(self) -> Dict[str, Any]:
"""Get the parameters used for the openai client."""
return {**self._default_params}
[docs]
def completion_with_retry(
self, run_manager: Optional[CallbackManagerForLLMRun] = None, **kwargs: Any
) -> Any:
"""Use tenacity to retry the completion call."""
retry_decorator = _create_retry_decorator(self, run_manager=run_manager)
@retry_decorator
def _completion_with_retry(**kwargs: Any) -> Any:
try:
request_timeout = kwargs.pop("request_timeout")
request = Requests(headers=self._headers())
response = request.post(
url=self._url(), data=self._body(kwargs), timeout=request_timeout
)
self._handle_status(response.status_code, response.text)
return response
except Exception as e:
print("EX", e) # noqa: T201
raise
return _completion_with_retry(**kwargs)
[docs]
async def acompletion_with_retry(
self,
run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
**kwargs: Any,
) -> Any:
"""Use tenacity to retry the async completion call."""
retry_decorator = _create_retry_decorator(self, run_manager=run_manager)
@retry_decorator
async def _completion_with_retry(**kwargs: Any) -> Any:
try:
request_timeout = kwargs.pop("request_timeout")
request = Requests(headers=self._headers())
async with request.apost(
url=self._url(), data=self._body(kwargs), timeout=request_timeout
) as response:
self._handle_status(response.status, response.text)
return await response.json()
except Exception as e:
print("EX", e) # noqa: T201
raise
return await _completion_with_retry(**kwargs)
@model_validator(mode="before")
@classmethod
def init_defaults(cls, values: Dict) -> Any:
"""Validate api key, python package exists, temperature, top_p, and top_k."""
# For compatibility with LiteLLM
api_key = get_from_dict_or_env(
values,
"deepinfra_api_key",
"DEEPINFRA_API_KEY",
default="",
)
values["deepinfra_api_token"] = get_from_dict_or_env(
values,
"deepinfra_api_token",
"DEEPINFRA_API_TOKEN",
default=api_key,
)
return values
@model_validator(mode="after")
def validate_environment(self) -> Self:
if self.temperature is not None and not 0 <= self.temperature <= 1:
raise ValueError("temperature must be in the range [0.0, 1.0]")
if self.top_p is not None and not 0 <= self.top_p <= 1:
raise ValueError("top_p must be in the range [0.0, 1.0]")
if self.top_k is not None and self.top_k <= 0:
raise ValueError("top_k must be positive")
return self
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, **kwargs}
response = self.completion_with_retry(
messages=message_dicts, run_manager=run_manager, **params
)
return self._create_chat_result(response.json())
def _create_chat_result(self, response: Mapping[str, Any]) -> ChatResult:
generations = []
for res in response["choices"]:
message = _convert_dict_to_message(res["message"])
gen = ChatGeneration(
message=message,
generation_info=dict(finish_reason=res.get("finish_reason")),
)
generations.append(gen)
token_usage = response.get("usage", {})
llm_output = {"token_usage": token_usage, "model": self.model_name}
res = ChatResult(generations=generations, llm_output=llm_output)
return res
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 _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}
response = self.completion_with_retry(
messages=message_dicts, run_manager=run_manager, **params
)
for line in _parse_stream(response.iter_lines()):
chunk = _handle_sse_line(line)
if chunk:
cg_chunk = ChatGenerationChunk(message=chunk, generation_info=None)
if run_manager:
run_manager.on_llm_new_token(str(chunk.content), chunk=cg_chunk)
yield cg_chunk
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 = {"messages": message_dicts, "stream": True, **params, **kwargs}
request_timeout = params.pop("request_timeout")
request = Requests(headers=self._headers())
async with request.apost(
url=self._url(), data=self._body(params), timeout=request_timeout
) as response:
async for line in _parse_stream_async(response.content):
chunk = _handle_sse_line(line)
if chunk:
cg_chunk = ChatGenerationChunk(message=chunk, generation_info=None)
if run_manager:
await run_manager.on_llm_new_token(
str(chunk.content), 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 = {"messages": message_dicts, **params, **kwargs}
res = await self.acompletion_with_retry(run_manager=run_manager, **params)
return self._create_chat_result(res)
@property
def _identifying_params(self) -> Dict[str, Any]:
"""Get the identifying parameters."""
return {
"model": self.model_name,
"temperature": self.temperature,
"top_p": self.top_p,
"top_k": self.top_k,
"n": self.n,
}
@property
def _llm_type(self) -> str:
return "deepinfra-chat"
def _handle_status(self, code: int, text: Any) -> None:
if code >= 500:
raise ChatDeepInfraException(
f"DeepInfra Server error status {code}: {text}"
)
elif code >= 400:
raise ValueError(f"DeepInfra received an invalid payload: {text}")
elif code != 200:
raise Exception(
f"DeepInfra returned an unexpected response with status "
f"{code}: {text}"
)
def _url(self) -> str:
return self.url
def _headers(self) -> Dict:
return {
"Authorization": f"bearer {self.deepinfra_api_token}",
"Content-Type": "application/json",
}
def _body(self, kwargs: Any) -> Dict:
return kwargs
def _parse_stream(rbody: Iterator[bytes]) -> Iterator[str]:
for line in rbody:
_line = _parse_stream_helper(line)
if _line is not None:
yield _line
async def _parse_stream_async(rbody: aiohttp.StreamReader) -> AsyncIterator[str]:
async for line in rbody:
_line = _parse_stream_helper(line)
if _line is not None:
yield _line
def _parse_stream_helper(line: bytes) -> Optional[str]:
if line and line.startswith(b"data:"):
if line.startswith(b"data: "):
# SSE event may be valid when it contain whitespace
line = line[len(b"data: ") :]
else:
line = line[len(b"data:") :]
if line.strip() == b"[DONE]":
# return here will cause GeneratorExit exception in urllib3
# and it will close http connection with TCP Reset
return None
else:
return line.decode("utf-8")
return None
def _handle_sse_line(line: str) -> Optional[BaseMessageChunk]:
try:
obj = json.loads(line)
default_chunk_class = AIMessageChunk
delta = obj.get("choices", [{}])[0].get("delta", {})
return _convert_delta_to_message_chunk(delta, default_chunk_class)
except Exception:
return None
