From c909cb78d8d9e55cbf48bb473a8a6b4d5b2f325c Mon Sep 17 00:00:00 2001 From: ulleo Date: Thu, 10 Jul 2025 19:13:57 +0800 Subject: [PATCH] feat: add chart reasoning content --- backend/apps/ai_model/model_factory.py | 64 ++---- backend/apps/ai_model/openai/llm.py | 168 +++++++++++++- backend/apps/chat/api/chat.py | 35 ++- backend/apps/chat/curd/chat.py | 79 ++++++- backend/apps/chat/models/chat_model.py | 4 +- backend/apps/chat/task/llm.py | 208 +++++++++++------- backend/template.yaml | 30 +-- frontend/src/api/chat.ts | 30 ++- frontend/src/entity/supplier.ts | 2 +- frontend/src/views/chat/ChatAnswer.vue | 6 +- .../chat/component/PredictChartBlock.vue | 33 ++- frontend/src/views/chat/index.vue | 118 +++++++++- 12 files changed, 592 insertions(+), 185 deletions(-) diff --git a/backend/apps/ai_model/model_factory.py b/backend/apps/ai_model/model_factory.py index 605c1d03..7a65f873 100644 --- a/backend/apps/ai_model/model_factory.py +++ b/backend/apps/ai_model/model_factory.py @@ -1,13 +1,14 @@ import json -from pydantic import BaseModel -from typing import Optional, Dict, Any, Type from abc import ABC, abstractmethod -from langchain_core.language_models import BaseLLM as LangchainBaseLLM -from langchain_openai import ChatOpenAI +from typing import Optional, Dict, Any, Type + +from langchain.chat_models.base import BaseChatModel +from pydantic import BaseModel from sqlmodel import Session, select -from common.core.db import engine +from apps.ai_model.openai.llm import BaseChatOpenAI from apps.system.models.system_model import AiModelDetail +from common.core.db import engine # from langchain_community.llms import Tongyi, VLLM @@ -20,76 +21,57 @@ class LLMConfig(BaseModel): api_key: Optional[str] = None api_base_url: Optional[str] = None additional_params: Dict[str, Any] = {} - + class BaseLLM(ABC): """Abstract base class for large language models""" - + def __init__(self, config: LLMConfig): self.config = config self._llm = self._init_llm() - + @abstractmethod - def _init_llm(self) -> LangchainBaseLLM: + def _init_llm(self) -> BaseChatModel: """Initialize specific large language model instance""" pass - + @property - def llm(self) -> LangchainBaseLLM: + def llm(self) -> BaseChatModel: """Return the langchain LLM instance""" return self._llm + class OpenAILLM(BaseLLM): - def _init_llm(self) -> LangchainBaseLLM: - return ChatOpenAI( + def _init_llm(self) -> BaseChatModel: + return BaseChatOpenAI( model=self.config.model_name, api_key=self.config.api_key, base_url=self.config.api_base_url, stream_usage=True, - **self.config.additional_params + **self.config.additional_params, + extra_body={"enable_thinking": True}, ) - + def generate(self, prompt: str) -> str: return self.llm.invoke(prompt) -""" class TongyiLLM(BaseLLM): - def _init_llm(self) -> LangchainBaseLLM: - return Tongyi( - model_name=self.config.model_name, - dashscope_api_key=self.config.api_key, - **self.config.additional_params - ) - - def generate(self, prompt: str) -> str: - return self.llm.invoke(prompt) - -class VLLMLLM(BaseLLM): - def _init_llm(self) -> LangchainBaseLLM: - return VLLM( - model=self.config.model_name, - **self.config.additional_params - ) - - def generate(self, prompt: str) -> str: - return self.llm.invoke(prompt) """ - class LLMFactory: """Large Language Model Factory Class""" - + _llm_types: Dict[str, Type[BaseLLM]] = { "openai": OpenAILLM, "tongyi": OpenAILLM, "vllm": OpenAILLM } - + @classmethod def create_llm(cls, config: LLMConfig) -> BaseLLM: llm_class = cls._llm_types.get(config.model_type) if not llm_class: raise ValueError(f"Unsupported LLM type: {config.model_type}") return llm_class(config) - + @classmethod def register_llm(cls, model_type: str, llm_class: Type[BaseLLM]): """Register new model type""" @@ -107,6 +89,7 @@ class LLMFactory: ) return config """ + def get_default_config() -> LLMConfig: with Session(engine) as session: db_model = session.exec( @@ -130,6 +113,5 @@ def get_default_config() -> LLMConfig: model_name=db_model.base_model, api_key=db_model.api_key, api_base_url=db_model.api_domain, - additional_params=additional_params + additional_params=additional_params, ) - \ No newline at end of file diff --git a/backend/apps/ai_model/openai/llm.py b/backend/apps/ai_model/openai/llm.py index 044a4824..2224868d 100644 --- a/backend/apps/ai_model/openai/llm.py +++ b/backend/apps/ai_model/openai/llm.py @@ -1 +1,167 @@ -# todo +from typing import Dict, Optional, Any, Iterator, cast, Mapping + +from langchain_core.language_models import LanguageModelInput +from langchain_core.messages import BaseMessage, BaseMessageChunk, HumanMessageChunk, AIMessageChunk, \ + SystemMessageChunk, FunctionMessageChunk, ChatMessageChunk +from langchain_core.messages.ai import UsageMetadata +from langchain_core.messages.tool import tool_call_chunk, ToolMessageChunk +from langchain_core.outputs import ChatGenerationChunk +from langchain_core.runnables import RunnableConfig, ensure_config +from langchain_openai import ChatOpenAI +from langchain_openai.chat_models.base import _create_usage_metadata + + +def _convert_delta_to_message_chunk( + _dict: Mapping[str, Any], default_class: type[BaseMessageChunk] +) -> BaseMessageChunk: + id_ = _dict.get("id") + role = cast(str, _dict.get("role")) + content = cast(str, _dict.get("content") or "") + additional_kwargs: dict = {} + if 'reasoning_content' in _dict: + additional_kwargs['reasoning_content'] = _dict.get('reasoning_content') + 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 + tool_call_chunks = [] + if raw_tool_calls := _dict.get("tool_calls"): + additional_kwargs["tool_calls"] = raw_tool_calls + try: + tool_call_chunks = [ + tool_call_chunk( + name=rtc["function"].get("name"), + args=rtc["function"].get("arguments"), + id=rtc.get("id"), + index=rtc["index"], + ) + for rtc in raw_tool_calls + ] + except KeyError: + pass + + if role == "user" or default_class == HumanMessageChunk: + return HumanMessageChunk(content=content, id=id_) + elif role == "assistant" or default_class == AIMessageChunk: + return AIMessageChunk( + content=content, + additional_kwargs=additional_kwargs, + id=id_, + tool_call_chunks=tool_call_chunks, # type: ignore[arg-type] + ) + elif role in ("system", "developer") or default_class == SystemMessageChunk: + if role == "developer": + additional_kwargs = {"__openai_role__": "developer"} + else: + additional_kwargs = {} + return SystemMessageChunk( + content=content, id=id_, additional_kwargs=additional_kwargs + ) + elif role == "function" or default_class == FunctionMessageChunk: + return FunctionMessageChunk(content=content, name=_dict["name"], id=id_) + elif role == "tool" or default_class == ToolMessageChunk: + return ToolMessageChunk( + content=content, tool_call_id=_dict["tool_call_id"], id=id_ + ) + elif role or default_class == ChatMessageChunk: + return ChatMessageChunk(content=content, role=role, id=id_) + else: + return default_class(content=content, id=id_) + + +class BaseChatOpenAI(ChatOpenAI): + usage_metadata: dict = {} + + # custom_get_token_ids = custom_get_token_ids + + def get_last_generation_info(self) -> Optional[Dict[str, Any]]: + return self.usage_metadata + + def _stream(self, *args: Any, **kwargs: Any) -> Iterator[ChatGenerationChunk]: + kwargs['stream_usage'] = True + for chunk in super()._stream(*args, **kwargs): + if chunk.message.usage_metadata is not None: + self.usage_metadata = chunk.message.usage_metadata + yield chunk + + def _convert_chunk_to_generation_chunk( + self, + chunk: dict, + default_chunk_class: type, + base_generation_info: Optional[dict], + ) -> Optional[ChatGenerationChunk]: + if chunk.get("type") == "content.delta": # from beta.chat.completions.stream + return None + token_usage = chunk.get("usage") + choices = ( + chunk.get("choices", []) + # from beta.chat.completions.stream + or chunk.get("chunk", {}).get("choices", []) + ) + + usage_metadata: Optional[UsageMetadata] = ( + _create_usage_metadata(token_usage) if token_usage and token_usage.get("prompt_tokens") else None + ) + if len(choices) == 0: + # logprobs is implicitly None + generation_chunk = ChatGenerationChunk( + message=default_chunk_class(content="", usage_metadata=usage_metadata) + ) + return generation_chunk + + choice = choices[0] + if choice["delta"] is None: + return None + + message_chunk = _convert_delta_to_message_chunk( + choice["delta"], default_chunk_class + ) + generation_info = {**base_generation_info} if base_generation_info else {} + + if finish_reason := choice.get("finish_reason"): + generation_info["finish_reason"] = finish_reason + if model_name := chunk.get("model"): + generation_info["model_name"] = model_name + if system_fingerprint := chunk.get("system_fingerprint"): + generation_info["system_fingerprint"] = system_fingerprint + + logprobs = choice.get("logprobs") + if logprobs: + generation_info["logprobs"] = logprobs + + if usage_metadata and isinstance(message_chunk, AIMessageChunk): + message_chunk.usage_metadata = usage_metadata + + generation_chunk = ChatGenerationChunk( + message=message_chunk, generation_info=generation_info or None + ) + return generation_chunk + + def invoke( + self, + input: LanguageModelInput, + config: Optional[RunnableConfig] = None, + *, + stop: Optional[list[str]] = None, + **kwargs: Any, + ) -> BaseMessage: + config = ensure_config(config) + chat_result = cast( + "ChatGeneration", + self.generate_prompt( + [self._convert_input(input)], + stop=stop, + callbacks=config.get("callbacks"), + tags=config.get("tags"), + metadata=config.get("metadata"), + run_name=config.get("run_name"), + run_id=config.pop("run_id", None), + **kwargs, + ).generations[0][0], + + ).message + + self.usage_metadata = chat_result.response_metadata[ + 'token_usage'] if 'token_usage' in chat_result.response_metadata else chat_result.usage_metadata + return chat_result diff --git a/backend/apps/chat/api/chat.py b/backend/apps/chat/api/chat.py index 18a2f342..4044d898 100644 --- a/backend/apps/chat/api/chat.py +++ b/backend/apps/chat/api/chat.py @@ -4,7 +4,7 @@ from fastapi import APIRouter, HTTPException from fastapi.responses import StreamingResponse from apps.chat.curd.chat import list_chats, get_chat_with_records, create_chat, rename_chat, \ - delete_chat + delete_chat, get_chat_chart_data, get_chat_predict_data from apps.chat.models.chat_model import CreateChat, ChatRecord, RenameChat, ChatQuestion from apps.chat.task.llm import LLMService, run_task, run_analysis_or_predict_task, run_recommend_questions_task from common.core.deps import SessionDep, CurrentUser @@ -28,6 +28,28 @@ async def get_chat(session: SessionDep, current_user: CurrentUser, chart_id: int ) +@router.get("/record/get/{chart_record_id}/data") +async def chat_record_data(session: SessionDep, chart_record_id: int): + try: + return get_chat_chart_data(chart_record_id=chart_record_id, session=session) + except Exception as e: + raise HTTPException( + status_code=500, + detail=str(e) + ) + + +@router.get("/record/get/{chart_record_id}/predict_data") +async def chat_predict_data(session: SessionDep, chart_record_id: int): + try: + return get_chat_predict_data(chart_record_id=chart_record_id, session=session) + except Exception as e: + raise HTTPException( + status_code=500, + detail=str(e) + ) + + @router.post("/rename") async def rename(session: SessionDep, chat: RenameChat): try: @@ -61,7 +83,7 @@ async def start_chat(session: SessionDep, current_user: CurrentUser, create_chat ) -@router.get("/recommend_questions/{chat_record_id}") +@router.post("/recommend_questions/{chat_record_id}") async def recommend_questions(session: SessionDep, current_user: CurrentUser, chat_record_id: int): try: record = session.query(ChatRecord).get(chat_record_id) @@ -74,11 +96,14 @@ async def recommend_questions(session: SessionDep, current_user: CurrentUser, ch llm_service = LLMService(session, current_user, request_question) llm_service.set_record(record) - - return run_recommend_questions_task(llm_service) except Exception as e: traceback.print_exc() - return '[]' + raise HTTPException( + status_code=500, + detail=str(e) + ) + + return StreamingResponse(run_recommend_questions_task(llm_service), media_type="text/event-stream") @router.post("/question") diff --git a/backend/apps/chat/curd/chat.py b/backend/apps/chat/curd/chat.py index 409a8529..cd143ae0 100644 --- a/backend/apps/chat/curd/chat.py +++ b/backend/apps/chat/curd/chat.py @@ -1,7 +1,8 @@ import datetime from typing import List -from sqlalchemy import and_, distinct +import orjson +from sqlalchemy import and_ from sqlalchemy.orm import load_only from apps.chat.models.chat_model import Chat, ChatRecord, CreateChat, ChatInfo, RenameChat, ChatQuestion @@ -42,6 +43,26 @@ def delete_chat(session, chart_id) -> str: return f'Chat with id {chart_id} has been deleted' +def get_chat_chart_data(session: SessionDep, chart_record_id: int): + res = session.query(ChatRecord).options(load_only(ChatRecord.data)).get(chart_record_id) + if res: + try: + return orjson.loads(res.data) + except Exception: + pass + return {} + + +def get_chat_predict_data(session: SessionDep, chart_record_id: int): + res = session.query(ChatRecord).options(load_only(ChatRecord.predict_data)).get(chart_record_id) + if res: + try: + return orjson.loads(res.data) + except Exception: + pass + return '' + + def get_chat_with_records(session: SessionDep, chart_id: int, current_user: CurrentUser) -> ChatInfo: chat = session.query(Chat).filter(Chat.id == chart_id).first() if not chat: @@ -61,16 +82,54 @@ def get_chat_with_records(session: SessionDep, chart_id: int, current_user: Curr load_only(ChatRecord.id, ChatRecord.chat_id, ChatRecord.create_time, ChatRecord.finish_time, ChatRecord.question, ChatRecord.sql_answer, ChatRecord.sql, ChatRecord.data, ChatRecord.chart_answer, ChatRecord.chart, ChatRecord.analysis, ChatRecord.predict, - ChatRecord.datasource_select_answer, ChatRecord.recommended_question_answer, + ChatRecord.datasource_select_answer, ChatRecord.recommended_question, ChatRecord.first_chat, ChatRecord.predict_data, ChatRecord.finish, ChatRecord.error, ChatRecord.run_time)).filter( and_(Chat.create_by == current_user.id, ChatRecord.chat_id == chart_id)).order_by(ChatRecord.create_time).all() - chat_info.records = record_list + result = list(map(format_record, record_list)) + + chat_info.records = result return chat_info +def format_record(record: ChatRecord): + _dict = record.model_dump() + + if record.sql_answer and record.sql_answer.strip() != '' and record.sql_answer.strip()[0] == '{' and \ + record.sql_answer.strip()[-1] == '}': + _obj = orjson.loads(record.sql_answer) + _dict['sql_answer'] = _obj.get('reasoning_content') + if record.chart_answer and record.chart_answer.strip() != '' and record.chart_answer.strip()[0] == '{' and \ + record.chart_answer.strip()[-1] == '}': + _obj = orjson.loads(record.chart_answer) + _dict['chart_answer'] = _obj.get('reasoning_content') + if record.analysis and record.analysis.strip() != '' and record.analysis.strip()[0] == '{' and \ + record.analysis.strip()[-1] == '}': + _obj = orjson.loads(record.analysis) + _dict['analysis_thinking'] = _obj.get('reasoning_content') + _dict['analysis'] = _obj.get('content') + if record.predict and record.predict.strip() != '' and record.predict.strip()[0] == '{' and record.predict.strip()[ + -1] == '}': + _obj = orjson.loads(record.predict) + _dict['predict'] = _obj.get('reasoning_content') + _dict['predict_content'] = _obj.get('content') + if record.data and record.data.strip() != '': + try: + _obj = orjson.loads(record.data) + _dict['data'] = _obj + except Exception: + pass + if record.predict_data and record.predict_data.strip() != '': + try: + _obj = orjson.loads(record.predict_data) + _dict['predict_data'] = _obj + except Exception: + pass + return _dict + + def list_records(session: SessionDep, chart_id: int, current_user: CurrentUser) -> List[ChatRecord]: record_list = session.query(ChatRecord).filter( and_(Chat.create_by == current_user.id, ChatRecord.chat_id == chart_id)).order_by(ChatRecord.create_time).all() @@ -251,18 +310,18 @@ def save_full_select_datasource_message_and_answer(session: SessionDep, record_i return result -def save_full_recommend_question_message_and_answer(session: SessionDep, record_id: int, answer: str, - full_message: str) -> ChatRecord: +def save_full_recommend_question_message_and_answer(session: SessionDep, record_id: int, answer: dict = None, + full_message: str = '[]') -> ChatRecord: if not record_id: raise Exception("Record id cannot be None") record = session.query(ChatRecord).filter(ChatRecord.id == record_id).first() record.full_recommended_question_message = full_message - record.recommended_question_answer = answer + record.recommended_question_answer = orjson.dumps(answer).decode() json_str = '[]' - if answer and answer != '': + if answer and answer.get('content') and answer.get('content') != '': try: - json_str = extract_nested_json(answer) + json_str = extract_nested_json(answer.get('content')) if not json_str: json_str = '[]' @@ -339,7 +398,7 @@ def save_chart(session: SessionDep, record_id: int, chart: str) -> ChatRecord: return result -def save_predict_data(session: SessionDep, record_id: int, data: str) -> ChatRecord: +def save_predict_data(session: SessionDep, record_id: int, data: str = '') -> ChatRecord: if not record_id: raise Exception("Record id cannot be None") record = session.query(ChatRecord).filter(ChatRecord.id == record_id).first() @@ -414,6 +473,6 @@ def get_old_questions(session: SessionDep, datasource: int): if not datasource: return [] records = session.query(ChatRecord.question, ChatRecord.create_time).filter(ChatRecord.datasource == datasource, - ChatRecord.question != None).order_by( + ChatRecord.question != None).order_by( ChatRecord.create_time.desc()).limit(20).all() return records diff --git a/backend/apps/chat/models/chat_model.py b/backend/apps/chat/models/chat_model.py index 2de01ba4..9502a21d 100644 --- a/backend/apps/chat/models/chat_model.py +++ b/backend/apps/chat/models/chat_model.py @@ -2,7 +2,7 @@ from datetime import datetime from typing import List, Optional from pydantic import BaseModel -from sqlalchemy import Column, Text, BigInteger, DateTime, Integer, Identity, Boolean +from sqlalchemy import Column, Text, BigInteger, DateTime, Identity, Boolean from sqlmodel import SQLModel, Field from apps.template.generate_analysis.generator import get_analysis_template @@ -86,7 +86,7 @@ class ChatInfo(BaseModel): engine_type: str = '' datasource_name: str = '' datasource_exists: bool = True - records: List[ChatRecord] = [] + records: List[ChatRecord | dict] = [] class AiModelQuestion(BaseModel): diff --git a/backend/apps/chat/task/llm.py b/backend/apps/chat/task/llm.py index 78a84850..6c94d05f 100644 --- a/backend/apps/chat/task/llm.py +++ b/backend/apps/chat/task/llm.py @@ -7,9 +7,9 @@ import numpy as np import orjson import pandas as pd import requests +from langchain.chat_models.base import BaseChatModel from langchain_community.utilities import SQLDatabase -from langchain_core.language_models import BaseLLM -from langchain_core.messages import BaseMessage, SystemMessage, HumanMessage, AIMessage, AIMessageChunk +from langchain_core.messages import BaseMessage, SystemMessage, HumanMessage, AIMessage from sqlalchemy import select from sqlalchemy.orm import load_only @@ -37,7 +37,7 @@ class LLMService: chat_question: ChatQuestion record: ChatRecord config: LLMConfig - llm: BaseLLM + llm: BaseChatModel sql_message: List[Union[BaseMessage, dict[str, Any]]] = [] chart_message: List[Union[BaseMessage, dict[str, Any]]] = [] history_records: List[ChatRecord] = [] @@ -195,23 +195,27 @@ class LLMService: 'content': msg.content} for msg in analysis_msg]).decode()) - + full_thinking_text = '' full_analysis_text = '' res = self.llm.stream(analysis_msg) for chunk in res: print(chunk) - if isinstance(chunk, dict): - full_analysis_text += chunk['content'] - yield chunk['content'] - continue - if isinstance(chunk, AIMessageChunk): - full_analysis_text += chunk.content - yield chunk.content - continue + reasoning_content_chunk = '' + if 'reasoning_content' in chunk.additional_kwargs: + reasoning_content_chunk = chunk.additional_kwargs.get('reasoning_content', '') + # else: + # reasoning_content_chunk = chunk.get('reasoning_content') + if reasoning_content_chunk is None: + reasoning_content_chunk = '' + full_thinking_text += reasoning_content_chunk + + full_analysis_text += chunk.content + yield {'content': chunk.content, 'reasoning_content': reasoning_content_chunk} analysis_msg.append(AIMessage(full_analysis_text)) self.record = save_full_analysis_message_and_answer(session=self.session, record_id=self.record.id, - answer=full_analysis_text, + answer=orjson.dumps({'content': full_analysis_text, + 'reasoning_content': full_thinking_text}).decode(), full_message=orjson.dumps(history_msg + [{'type': msg.type, 'content': msg.content} for msg @@ -238,23 +242,28 @@ class LLMService: 'content': msg.content} for msg in predict_msg]).decode()) - + full_thinking_text = '' full_predict_text = '' res = self.llm.stream(predict_msg) for chunk in res: print(chunk) - if isinstance(chunk, dict): - full_predict_text += chunk['content'] - yield chunk['content'] - continue - if isinstance(chunk, AIMessageChunk): - full_predict_text += chunk.content - yield chunk.content - continue + reasoning_content_chunk = '' + if 'reasoning_content' in chunk.additional_kwargs: + reasoning_content_chunk = chunk.additional_kwargs.get('reasoning_content', '') + # else: + # reasoning_content_chunk = chunk.get('reasoning_content') + if reasoning_content_chunk is None: + reasoning_content_chunk = '' + full_thinking_text += reasoning_content_chunk + + full_predict_text += chunk.content + yield {'content': chunk.content, 'reasoning_content': reasoning_content_chunk} predict_msg.append(AIMessage(full_predict_text)) self.record = save_full_predict_message_and_answer(session=self.session, record_id=self.record.id, - answer=full_predict_text, data='', + answer=orjson.dumps({'content': full_predict_text, + 'reasoning_content': full_thinking_text}).decode(), + data='', full_message=orjson.dumps(history_msg + [{'type': msg.type, 'content': msg.content} for msg @@ -269,38 +278,44 @@ class LLMService: guess_msg: List[Union[BaseMessage, dict[str, Any]]] = [] guess_msg.append(SystemMessage(content=self.chat_question.guess_sys_question())) - # todo old questions + old_questions = list(map(lambda q: q[0].strip(), get_old_questions(self.session, self.record.datasource))) - guess_msg.append(HumanMessage(content=self.chat_question.guess_user_question(orjson.dumps(old_questions).decode()))) + guess_msg.append( + HumanMessage(content=self.chat_question.guess_user_question(orjson.dumps(old_questions).decode()))) self.record = save_full_recommend_question_message_and_answer(session=self.session, record_id=self.record.id, - answer='', full_message=orjson.dumps([{'type': msg.type, 'content': msg.content} for msg in guess_msg]).decode()) - + full_thinking_text = '' full_guess_text = '' res = self.llm.stream(guess_msg) for chunk in res: print(chunk) - if isinstance(chunk, dict): - full_guess_text += chunk['content'] - continue - if isinstance(chunk, AIMessageChunk): - full_guess_text += chunk.content - continue + reasoning_content_chunk = '' + if 'reasoning_content' in chunk.additional_kwargs: + reasoning_content_chunk = chunk.additional_kwargs.get('reasoning_content', '') + # else: + # reasoning_content_chunk = chunk.get('reasoning_content') + if reasoning_content_chunk is None: + reasoning_content_chunk = '' + full_thinking_text += reasoning_content_chunk + + full_guess_text += chunk.content + yield {'content': chunk.content, 'reasoning_content': reasoning_content_chunk} guess_msg.append(AIMessage(full_guess_text)) self.record = save_full_recommend_question_message_and_answer(session=self.session, record_id=self.record.id, - answer=full_guess_text, + answer={'content': full_guess_text, + 'reasoning_content': full_thinking_text}, full_message=orjson.dumps([{'type': msg.type, 'content': msg.content} for msg in guess_msg]).decode()) - return self.record.recommended_question + yield {'recommended_question': self.record.recommended_question} def select_datasource(self): datasource_msg: List[Union[BaseMessage, dict[str, Any]]] = [] @@ -325,17 +340,22 @@ class LLMService: for msg in datasource_msg]).decode()) + full_thinking_text = '' full_text = '' res = self.llm.stream(datasource_msg) for chunk in res: - if isinstance(chunk, dict): - full_text += chunk['content'] - yield chunk['content'] - continue - if isinstance(chunk, AIMessageChunk): - full_text += chunk.content - yield chunk.content - continue + print(chunk) + reasoning_content_chunk = '' + if 'reasoning_content' in chunk.additional_kwargs: + reasoning_content_chunk = chunk.additional_kwargs.get('reasoning_content', '') + # else: + # reasoning_content_chunk = chunk.get('reasoning_content') + if reasoning_content_chunk is None: + reasoning_content_chunk = '' + full_thinking_text += reasoning_content_chunk + + full_text += chunk.content + yield {'content': chunk.content, 'reasoning_content': reasoning_content_chunk} datasource_msg.append(AIMessage(full_text)) json_str = extract_nested_json(full_text) @@ -374,7 +394,9 @@ class LLMService: _error = e self.record = save_full_select_datasource_message_and_answer(session=self.session, record_id=self.record.id, - answer=full_text, datasource=_datasource, + answer=orjson.dumps({'content': full_text, + 'reasoning_content': full_thinking_text}).decode(), + datasource=_datasource, engine_type=_engine_type, full_message=orjson.dumps(history_msg + [{'type': msg.type, @@ -394,22 +416,27 @@ class LLMService: full_message=orjson.dumps( [{'type': msg.type, 'content': msg.content} for msg in self.sql_message]).decode()) + full_thinking_text = '' full_sql_text = '' res = self.llm.stream(self.sql_message) for chunk in res: print(chunk) - if isinstance(chunk, dict): - full_sql_text += chunk['content'] - yield chunk['content'] - continue - if isinstance(chunk, AIMessageChunk): - full_sql_text += chunk.content - yield chunk.content - continue + reasoning_content_chunk = '' + if 'reasoning_content' in chunk.additional_kwargs: + reasoning_content_chunk = chunk.additional_kwargs.get('reasoning_content', '') + # else: + # reasoning_content_chunk = chunk.get('reasoning_content') + if reasoning_content_chunk is None: + reasoning_content_chunk = '' + full_thinking_text += reasoning_content_chunk + + full_sql_text += chunk.content + yield {'content': chunk.content, 'reasoning_content': reasoning_content_chunk} self.sql_message.append(AIMessage(full_sql_text)) self.record = save_full_sql_message_and_answer(session=self.session, record_id=self.record.id, - answer=full_sql_text, + answer=orjson.dumps({'content': full_sql_text, + 'reasoning_content': full_thinking_text}).decode(), full_message=orjson.dumps( [{'type': msg.type, 'content': msg.content} for msg in self.sql_message]).decode()) @@ -421,21 +448,27 @@ class LLMService: full_message=orjson.dumps( [{'type': msg.type, 'content': msg.content} for msg in self.chart_message]).decode()) + full_thinking_text = '' full_chart_text = '' res = self.llm.stream(self.chart_message) for chunk in res: - if isinstance(chunk, dict): - full_chart_text += chunk['content'] - yield chunk['content'] - continue - if isinstance(chunk, AIMessageChunk): - full_chart_text += chunk.content - yield chunk.content - continue + print(chunk) + reasoning_content_chunk = '' + if 'reasoning_content' in chunk.additional_kwargs: + reasoning_content_chunk = chunk.additional_kwargs.get('reasoning_content', '') + # else: + # reasoning_content_chunk = chunk.get('reasoning_content') + if reasoning_content_chunk is None: + reasoning_content_chunk = '' + full_thinking_text += reasoning_content_chunk + + full_chart_text += chunk.content + yield {'content': chunk.content, 'reasoning_content': reasoning_content_chunk} self.chart_message.append(AIMessage(full_chart_text)) self.record = save_full_chart_message_and_answer(session=self.session, record_id=self.record.id, - answer=full_chart_text, + answer=orjson.dumps({'content': full_chart_text, + 'reasoning_content': full_thinking_text}).decode(), full_message=orjson.dumps( [{'type': msg.type, 'content': msg.content} for msg in self.chart_message]).decode()) @@ -492,7 +525,7 @@ class LLMService: return chart - def check_save_predict_data(self, res: str) -> Dict[str, Any]: + def check_save_predict_data(self, res: str) -> bool: json_str = extract_nested_json(res) @@ -501,7 +534,10 @@ class LLMService: save_predict_data(session=self.session, record_id=self.record.id, data=json_str) - return json_str + if json_str == '': + return False + + return True def save_error(self, message: str): return save_error_message(session=self.session, record_id=self.record.id, message=message) @@ -566,7 +602,9 @@ def run_task(llm_service: LLMService, session: SessionDep, in_chat: bool = True) for chunk in ds_res: print(chunk) if in_chat: - yield orjson.dumps({'content': chunk, 'type': 'datasource-result'}).decode() + '\n\n' + yield orjson.dumps( + {'content': chunk.get('content'), 'reasoning_content': chunk.get('reasoning_content'), + 'type': 'datasource-result'}).decode() + '\n\n' if in_chat: yield orjson.dumps({'id': llm_service.ds.id, 'datasource_name': llm_service.ds.name, 'engine_type': llm_service.ds.type_name, 'type': 'datasource'}).decode() + '\n\n' @@ -577,9 +615,11 @@ def run_task(llm_service: LLMService, session: SessionDep, in_chat: bool = True) sql_res = llm_service.generate_sql() full_sql_text = '' for chunk in sql_res: - full_sql_text += chunk + full_sql_text += chunk.get('content') if in_chat: - yield orjson.dumps({'content': chunk, 'type': 'sql-result'}).decode() + '\n\n' + yield orjson.dumps( + {'content': chunk.get('content'), 'reasoning_content': chunk.get('reasoning_content'), + 'type': 'sql-result'}).decode() + '\n\n' if in_chat: yield orjson.dumps({'type': 'info', 'msg': 'sql generated'}).decode() + '\n\n' @@ -596,15 +636,17 @@ def run_task(llm_service: LLMService, session: SessionDep, in_chat: bool = True) result = llm_service.execute_sql(sql=sql) llm_service.save_sql_data(data_obj=result) if in_chat: - yield orjson.dumps({'content': orjson.dumps(result).decode(), 'type': 'sql-data'}).decode() + '\n\n' + yield orjson.dumps({'content': 'execute-success', 'type': 'sql-data'}).decode() + '\n\n' # generate chart chart_res = llm_service.generate_chart() full_chart_text = '' for chunk in chart_res: - full_chart_text += chunk + full_chart_text += chunk.get('content') if in_chat: - yield orjson.dumps({'content': chunk, 'type': 'chart-result'}).decode() + '\n\n' + yield orjson.dumps( + {'content': chunk.get('content'), 'reasoning_content': chunk.get('reasoning_content'), + 'type': 'chart-result'}).decode() + '\n\n' if in_chat: yield orjson.dumps({'type': 'info', 'msg': 'chart generated'}).decode() + '\n\n' @@ -667,7 +709,9 @@ def run_analysis_or_predict_task(llm_service: LLMService, action_type: str): # generate analysis analysis_res = llm_service.generate_analysis() for chunk in analysis_res: - yield orjson.dumps({'content': chunk, 'type': 'analysis-result'}).decode() + '\n\n' + yield orjson.dumps( + {'content': chunk.get('content'), 'reasoning_content': chunk.get('reasoning_content'), + 'type': 'analysis-result'}).decode() + '\n\n' yield orjson.dumps({'type': 'info', 'msg': 'analysis generated'}).decode() + '\n\n' yield orjson.dumps({'type': 'analysis_finish'}).decode() + '\n\n' @@ -677,12 +721,17 @@ def run_analysis_or_predict_task(llm_service: LLMService, action_type: str): analysis_res = llm_service.generate_predict() full_text = '' for chunk in analysis_res: - yield orjson.dumps({'content': chunk, 'type': 'predict-result'}).decode() + '\n\n' - full_text += chunk + yield orjson.dumps( + {'content': chunk.get('content'), 'reasoning_content': chunk.get('reasoning_content'), + 'type': 'predict-result'}).decode() + '\n\n' + full_text += chunk.get('content') yield orjson.dumps({'type': 'info', 'msg': 'predict generated'}).decode() + '\n\n' _data = llm_service.check_save_predict_data(res=full_text) - yield orjson.dumps({'type': 'predict', 'content': _data}).decode() + '\n\n' + if _data: + yield orjson.dumps({'type': 'predict-success'}).decode() + '\n\n' + else: + yield orjson.dumps({'type': 'predict-failed'}).decode() + '\n\n' yield orjson.dumps({'type': 'predict_finish'}).decode() + '\n\n' @@ -694,7 +743,16 @@ def run_analysis_or_predict_task(llm_service: LLMService, action_type: str): def run_recommend_questions_task(llm_service: LLMService): - return llm_service.generate_recommend_questions_task() + res = llm_service.generate_recommend_questions_task() + + for chunk in res: + if chunk.get('recommended_question'): + yield orjson.dumps( + {'content': chunk.get('recommended_question'), 'type': 'recommended_question'}).decode() + '\n\n' + else: + yield orjson.dumps( + {'content': chunk.get('content'), 'reasoning_content': chunk.get('reasoning_content'), + 'type': 'recommended_question_result'}).decode() + '\n\n' def request_picture(chat_id: int, record_id: int, chart: dict, data: dict): diff --git a/backend/template.yaml b/backend/template.yaml index f78677fd..a86e964a 100644 --- a/backend/template.yaml +++ b/backend/template.yaml @@ -23,13 +23,9 @@ template: - 如果问题是图表展示相关，可参考的图表类型为表格(table)、柱状图(column)、条形图(bar)、折线图(line)或饼图(pie)，返回的JSON： {{"success":true,"sql":"生成的SQL语句","chart-type":"选择的图表类型(table/column/bar/line/pie)"}} - ### 请遵守以下输出要求: - - 请逐步推理后给出答案 - - 推理过程的步骤中不需要输出任何JSON，仅在最终结果内输出符合要求的JSON - 步骤1: [思考内容] - 步骤2: [思考内容] - …… - 最终答案: [结果] + ### 响应, 请直接返回JSON结果: + ```json + user: | ### 表结构: {schema} @@ -37,7 +33,7 @@ template: ### 问题: {question} - ### 请使用 i18n: {lang} 对应的语言输出你的推理思考步骤、内容和结果 + ### 请使用 i18n: {lang} 对应的语言输出你的结果 ### 其他规则: {rule} @@ -69,17 +65,9 @@ template: 问题是：每个商品分类的平均价格 则生成的 JSON 可以是： {{"type":"table", "title": "每个商品分类的平均价格", "columns": [{{"name":"商品分类","value":"category"}}, {{"name":"平均价格","value":"average_price"}}]}} - ### 响应: - 根据您的指示，这是我生成的与 问题 和 sql 匹配的 JSON： + ### 响应, 请直接返回JSON结果: ```json - - ### 请遵守以下输出要求: - - 请逐步推理后给出答案 - - 推理过程的步骤中不需要输出任何JSON，仅在最终结果内输出符合要求的JSON - 步骤1: [思考内容] - 步骤2: [思考内容] - …… - 最终答案: [结果] + user: | ### SQL: {sql} @@ -87,7 +75,7 @@ template: ### 问题: {question} - ### 请使用 i18n: {lang} 对应的语言输出你的推理思考步骤、内容和结果 + ### 请使用 i18n: {lang} 对应的语言输出你的结果 ### 其他规则: {rule} @@ -108,8 +96,9 @@ template: - 最多返回4个你推测出的结果 - 若无法推测,则返回空数据JSON: [] + - 不需要思考过程，或者尽量精简思考过程 - ### 响应: + ### 响应, 请直接返回JSON结果: ```json user: | @@ -164,6 +153,7 @@ template: - 如果没有符合要求的数据源，则返回：{{"fail":"没有找到匹配的数据源"}} - 不需要思考过程，请直接返回JSON结果 + ### 响应, 请直接返回JSON结果: ```json user: | ### 请使用 i18n: {lang} 对应的语言输出你的结果 diff --git a/frontend/src/api/chat.ts b/frontend/src/api/chat.ts index aee7c42f..af085692 100644 --- a/frontend/src/api/chat.ts +++ b/frontend/src/api/chat.ts @@ -34,12 +34,14 @@ export class ChatRecord { question?: string sql_answer?: string sql?: string - data?: string + data?: string | any chart_answer?: string chart?: string analysis?: string + analysis_thinking?: string predict?: string - predict_data?: string + predict_content?: string + predict_data?: string | any finish?: boolean = false error?: string run_time: number = 0 @@ -55,12 +57,14 @@ export class ChatRecord { question: string, sql_answer: string | undefined, sql: string | undefined, - data: string | undefined, + data: string | any | undefined, chart_answer: string | undefined, chart: string | undefined, analysis: string | undefined, + analysis_thinking: string | undefined, predict: string | undefined, - predict_data: string | undefined, + predict_content: string | undefined, + predict_data: string | any | undefined, finish: boolean, error: string | undefined, run_time: number, @@ -75,12 +79,14 @@ export class ChatRecord { question?: string, sql_answer?: string, sql?: string, - data?: string, + data?: string | any, chart_answer?: string, chart?: string, analysis?: string, + analysis_thinking?: string, predict?: string, - predict_data?: string, + predict_content?: string, + predict_data?: string | any, finish?: boolean, error?: string, run_time?: number, @@ -98,7 +104,9 @@ export class ChatRecord { this.chart_answer = chart_answer this.chart = chart this.analysis = analysis + this.analysis_thinking = analysis_thinking this.predict = predict + this.predict_content = predict_content this.predict_data = predict_data this.finish = !!finish this.error = error @@ -219,7 +227,9 @@ const toChatRecord = (data?: any): ChatRecord | undefined => { data.chart_answer, data.chart, data.analysis, + data.analysis_thinking, data.predict, + data.predict_content, data.predict_data, data.finish, data.error, @@ -273,6 +283,12 @@ export const chatApi = { get: (id: number): Promise => { return request.get(`/chat/get/${id}`) }, + get_chart_data: (record_id: number): Promise => { + return request.get(`/chat/record/get/${record_id}/data`) + }, + get_chart_predict_data: (record_id: number): Promise => { + return request.get(`/chat/record/get/${record_id}/predict_data`) + }, startChat: (data: any): Promise => { return request.post('/chat/start', data) }, @@ -289,6 +305,6 @@ export const chatApi = { return request.fetchStream(`/chat/record/${record_id}/predict`, {}) }, recommendQuestions: (record_id: number | undefined) => { - return request.get(`/chat/recommend_questions/${record_id}`) + return request.fetchStream(`/chat/recommend_questions/${record_id}`, {}) }, } diff --git a/frontend/src/entity/supplier.ts b/frontend/src/entity/supplier.ts index 62a5c819..7e0a1aa0 100644 --- a/frontend/src/entity/supplier.ts +++ b/frontend/src/entity/supplier.ts @@ -38,7 +38,7 @@ export const supplierList: Array<{ model_config: { 0: { api_domain: 'https://api.deepseek.com', - model_options: ['deepseek-chat'], + model_options: ['deepseek-chat', 'deepseek-reasoner'], }, }, }, diff --git a/frontend/src/views/chat/ChatAnswer.vue b/frontend/src/views/chat/ChatAnswer.vue index e32a71e5..40a4e850 100644 --- a/frontend/src/views/chat/ChatAnswer.vue +++ b/frontend/src/views/chat/ChatAnswer.vue @@ -49,7 +49,11 @@ const dataObject = computed<{ data: Array<{ [key: string]: any }> }>(() => { if (props.message?.record?.data) { - return JSON.parse(props.message.record.data) + if (typeof props.message?.record?.data === 'string') { + return JSON.parse(props.message.record.data) + } else { + return props.message.record.data + } } return {} }) diff --git a/frontend/src/views/chat/component/PredictChartBlock.vue b/frontend/src/views/chat/component/PredictChartBlock.vue index 6d931481..a7290031 100644 --- a/frontend/src/views/chat/component/PredictChartBlock.vue +++ b/frontend/src/views/chat/component/PredictChartBlock.vue @@ -9,7 +9,7 @@ const props = defineProps<{ id?: number | string chartType: ChartTypes message: ChatMessage - data: string + data: string | any }>() const dataObject = computed<{ @@ -17,23 +17,32 @@ const dataObject = computed<{ data: Array<{ [key: string]: any }> }>(() => { if (props.message?.record?.data) { - return JSON.parse(props.message.record.data) + if (typeof props.message?.record?.data === 'string') { + return JSON.parse(props.message.record.data) + } else { + return props.message.record.data + } } return {} }) const _data = computed(() => { let _list = [] - if ( - props.data && - props.data.length > 0 && - props.data.trim().startsWith('[') && - props.data.trim().endsWith(']') - ) { - try { - _list = JSON.parse(props.data) - } catch (e) { - console.error(e) + if (props.data && typeof props.data === 'string') { + if ( + props.data.length > 0 && + props.data.trim().startsWith('[') && + props.data.trim().endsWith(']') + ) { + try { + _list = JSON.parse(props.data) + } catch (e) { + console.error(e) + } + } + } else { + if (props.data.length > 0) { + _list = props.data } } diff --git a/frontend/src/views/chat/index.vue b/frontend/src/views/chat/index.vue index 68c0c1a5..2b4be977 100644 --- a/frontend/src/views/chat/index.vue +++ b/frontend/src/views/chat/index.vue @@ -79,6 +79,7 @@ />