langchain/libs/community/langchain_community/chat_models/llama_edge.py

import json
import logging
import re
from typing import Any, Dict, Iterator, List, Mapping, Optional, Type

import requests
from langchain_core.callbacks import CallbackManagerForLLMRun
from langchain_core.language_models.chat_models import (
    BaseChatModel,
    generate_from_stream,
)
from langchain_core.messages import (
    AIMessage,
    AIMessageChunk,
    BaseMessage,
    BaseMessageChunk,
    ChatMessage,
    ChatMessageChunk,
    HumanMessage,
    HumanMessageChunk,
    SystemMessage,
)
from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, ChatResult
from langchain_core.pydantic_v1 import root_validator
from langchain_core.utils import get_pydantic_field_names

logger = logging.getLogger(__name__)


def _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
    role = _dict["role"]
    if role == "user":
        return HumanMessage(content=_dict["content"])
    elif role == "assistant":
        return AIMessage(content=_dict.get("content", "") or "")
    else:
        return ChatMessage(content=_dict["content"], role=role)


def _convert_message_to_dict(message: BaseMessage) -> dict:
    message_dict: Dict[str, Any]
    if isinstance(message, ChatMessage):
        message_dict = {"role": message.role, "content": message.content}
    elif isinstance(message, SystemMessage):
        message_dict = {"role": "system", "content": message.content}
    elif isinstance(message, HumanMessage):
        message_dict = {"role": "user", "content": message.content}
    elif isinstance(message, AIMessage):
        message_dict = {"role": "assistant", "content": message.content}
    else:
        raise TypeError(f"Got unknown type {message}")

    return message_dict


def _convert_delta_to_message_chunk(
    _dict: Mapping[str, Any], default_class: Type[BaseMessageChunk]
) -> BaseMessageChunk:
    role = _dict.get("role")
    content = _dict.get("content") or ""

    if role == "user" or default_class == HumanMessageChunk:
        return HumanMessageChunk(content=content)
    elif role == "assistant" or default_class == AIMessageChunk:
        return AIMessageChunk(content=content)
    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]


class LlamaEdgeChatService(BaseChatModel):
    """Chat with LLMs via `llama-api-server`

    For the information about `llama-api-server`, visit https://github.com/second-state/LlamaEdge
    """

    request_timeout: int = 60
    """request timeout for chat http requests"""
    service_url: Optional[str] = None
    """URL of WasmChat service"""
    model: str = "NA"
    """model name, default is `NA`."""
    streaming: bool = False
    """Whether to stream the results or not."""

    class Config:
        """Configuration for this pydantic object."""

        allow_population_by_field_name = True

    @root_validator(pre=True)
    def build_extra(cls, values: Dict[str, Any]) -> Dict[str, Any]:
        """Build extra kwargs from additional params that were passed in."""
        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

    def _generate(
        self,
        messages: List[BaseMessage],
        stop: Optional[List[str]] = None,
        run_manager: Optional[CallbackManagerForLLMRun] = None,
        **kwargs: Any,
    ) -> ChatResult:
        if self.streaming:
            stream_iter = self._stream(
                messages=messages, stop=stop, run_manager=run_manager, **kwargs
            )
            return generate_from_stream(stream_iter)

        res = self._chat(messages, **kwargs)

        if res.status_code != 200:
            raise ValueError(f"Error code: {res.status_code}, reason: {res.reason}")

        response = res.json()

        return self._create_chat_result(response)

    def _stream(
        self,
        messages: List[BaseMessage],
        stop: Optional[List[str]] = None,
        run_manager: Optional[CallbackManagerForLLMRun] = None,
        **kwargs: Any,
    ) -> Iterator[ChatGenerationChunk]:
        res = self._chat(messages, **kwargs)

        default_chunk_class = AIMessageChunk
        substring = '"object":"chat.completion.chunk"}'
        for line in res.iter_lines():
            chunks = []
            if line:
                json_string = line.decode("utf-8")

                # Find all positions of the substring
                positions = [m.start() for m in re.finditer(substring, json_string)]
                positions = [-1 * len(substring)] + positions

                for i in range(len(positions) - 1):
                    chunk = json.loads(
                        json_string[
                            positions[i] + len(substring) : positions[i + 1]
                            + len(substring)
                        ]
                    )
                    chunks.append(chunk)

            for chunk in chunks:
                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
                )
                if (
                    choice.get("finish_reason") is not None
                    and choice.get("finish_reason") == "stop"
                ):
                    break
                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 _chat(self, messages: List[BaseMessage], **kwargs: Any) -> requests.Response:
        if self.service_url is None:
            res = requests.models.Response()
            res.status_code = 503
            res.reason = "The IP address or port of the chat service is incorrect."
            return res

        service_url = f"{self.service_url}/v1/chat/completions"

        if self.streaming:
            payload = {
                "model": self.model,
                "messages": [_convert_message_to_dict(m) for m in messages],
                "stream": self.streaming,
            }
        else:
            payload = {
                "model": self.model,
                "messages": [_convert_message_to_dict(m) for m in messages],
            }

        res = requests.post(
            url=service_url,
            timeout=self.request_timeout,
            headers={
                "accept": "application/json",
                "Content-Type": "application/json",
            },
            data=json.dumps(payload),
        )

        return res

    def _create_chat_result(self, response: Mapping[str, Any]) -> ChatResult:
        message = _convert_dict_to_message(response["choices"][0].get("message"))
        generations = [ChatGeneration(message=message)]

        token_usage = response["usage"]
        llm_output = {"token_usage": token_usage, "model": self.model}
        return ChatResult(generations=generations, llm_output=llm_output)

    @property
    def _llm_type(self) -> str:
        return "wasm-chat"