LangBot/pkg/provider/modelmgr/requesters/chatcmpl.py

from __future__ import annotations

import asyncio
import typing

import openai
import openai.types.chat.chat_completion as chat_completion
import httpx

from .. import errors, requester
from ....core import entities as core_entities
from ... import entities as llm_entities
from ...tools import entities as tools_entities


class OpenAIChatCompletions(requester.ProviderAPIRequester):
    """OpenAI ChatCompletion API 请求器"""

    client: openai.AsyncClient

    default_config: dict[str, typing.Any] = {
        'base_url': 'https://api.openai.com/v1',
        'timeout': 120,
    }

    async def initialize(self):
        self.client = openai.AsyncClient(
            api_key='',
            base_url=self.requester_cfg['base_url'].replace(' ', ''),
            timeout=self.requester_cfg['timeout'],
            http_client=httpx.AsyncClient(trust_env=True, timeout=self.requester_cfg['timeout']),
        )

    async def _req(
        self,
        args: dict,
        extra_body: dict = {},
    ) -> chat_completion.ChatCompletion:
        return await self.client.chat.completions.create(**args, extra_body=extra_body)

    async def _make_msg(
        self,
        chat_completion: chat_completion.ChatCompletion,
    ) -> llm_entities.Message:
        chatcmpl_message = chat_completion.choices[0].message.model_dump()

        # 确保 role 字段存在且不为 None
        if 'role' not in chatcmpl_message or chatcmpl_message['role'] is None:
            chatcmpl_message['role'] = 'assistant'

        reasoning_content = chatcmpl_message['reasoning_content'] if 'reasoning_content' in chatcmpl_message else None

        # deepseek的reasoner模型
        if reasoning_content is not None:
            chatcmpl_message['content'] = '<think>\n' + reasoning_content + '\n</think>\n' + chatcmpl_message['content']

        message = llm_entities.Message(**chatcmpl_message)

        return message

    async def _closure(
        self,
        query: core_entities.Query,
        req_messages: list[dict],
        use_model: requester.RuntimeLLMModel,
        use_funcs: list[tools_entities.LLMFunction] = None,
        extra_args: dict[str, typing.Any] = {},
    ) -> llm_entities.Message:
        self.client.api_key = use_model.token_mgr.get_token()

        args = {}
        args['model'] = use_model.model_entity.name

        if use_funcs:
            tools = await self.ap.tool_mgr.generate_tools_for_openai(use_funcs)

            if tools:
                args['tools'] = tools

        # 设置此次请求中的messages
        messages = req_messages.copy()

        # 检查vision
        for msg in messages:
            if 'content' in msg and isinstance(msg['content'], list):
                for me in msg['content']:
                    if me['type'] == 'image_base64':
                        me['image_url'] = {'url': me['image_base64']}
                        me['type'] = 'image_url'
                        del me['image_base64']

        args['messages'] = messages

        # 发送请求
        resp = await self._req(args, extra_body=extra_args)

        # 处理请求结果
        message = await self._make_msg(resp)

        return message

    async def invoke_llm(
        self,
        query: core_entities.Query,
        model: requester.RuntimeLLMModel,
        messages: typing.List[llm_entities.Message],
        funcs: typing.List[tools_entities.LLMFunction] = None,
        extra_args: dict[str, typing.Any] = {},
    ) -> llm_entities.Message:
        req_messages = []  # req_messages 仅用于类内，外部同步由 query.messages 进行
        for m in messages:
            msg_dict = m.dict(exclude_none=True)
            content = msg_dict.get('content')
            if isinstance(content, list):
                # 检查 content 列表中是否每个部分都是文本
                if all(isinstance(part, dict) and part.get('type') == 'text' for part in content):
                    # 将所有文本部分合并为一个字符串
                    msg_dict['content'] = '\n'.join(part['text'] for part in content)
            req_messages.append(msg_dict)

        try:
            return await self._closure(
                query=query,
                req_messages=req_messages,
                use_model=model,
                use_funcs=funcs,
                extra_args=extra_args,
            )
        except asyncio.TimeoutError:
            raise errors.RequesterError('请求超时')
        except openai.BadRequestError as e:
            if 'context_length_exceeded' in e.message:
                raise errors.RequesterError(f'上文过长，请重置会话: {e.message}')
            else:
                raise errors.RequesterError(f'请求参数错误: {e.message}')
        except openai.AuthenticationError as e:
            raise errors.RequesterError(f'无效的 api-key: {e.message}')
        except openai.NotFoundError as e:
            raise errors.RequesterError(f'请求路径错误: {e.message}')
        except openai.RateLimitError as e:
            raise errors.RequesterError(f'请求过于频繁或余额不足: {e.message}')
        except openai.APIError as e:
            raise errors.RequesterError(f'请求错误: {e.message}')

    async def invoke_embedding(
        self,
        model: requester.RuntimeEmbeddingModel,
        input_text: list[str],
        extra_args: dict[str, typing.Any] = {},
    ) -> list[list[float]]:
        """调用 Embedding API"""
        self.client.api_key = model.token_mgr.get_token()

        args = {
            'model': model.model_entity.name,
            'input': input_text,
        }

        if model.model_entity.extra_args:
            args.update(model.model_entity.extra_args)

        args.update(extra_args)

        try:
            resp = await self.client.embeddings.create(**args)

            return [d.embedding for d in resp.data]
        except asyncio.TimeoutError:
            raise errors.RequesterError('请求超时')
        except openai.BadRequestError as e:
            raise errors.RequesterError(f'请求参数错误: {e.message}')
        except openai.AuthenticationError as e:
            raise errors.RequesterError(f'无效的 api-key: {e.message}')
        except openai.NotFoundError as e:
            raise errors.RequesterError(f'请求路径错误: {e.message}')
        except openai.RateLimitError as e:
            raise errors.RequesterError(f'请求过于频繁或余额不足: {e.message}')
        except openai.APIError as e:
            raise errors.RequesterError(f'请求错误: {e.message}')