from fastapi import Body, Request from sse_starlette.sse import EventSourceResponse from fastapi.concurrency import run_in_threadpool from configs import (LLM_MODELS, VECTOR_SEARCH_TOP_K, SCORE_THRESHOLD, TEMPERATURE, USE_RERANKER, RERANKER_MODEL, RERANKER_MAX_LENGTH, MODEL_PATH, MAX_TOKENS, MAX_CUT_TOKENS) from server.utils import wrap_done, get_ChatOpenAI from server.utils import BaseResponse, get_prompt_template, get_format_template from server.utils import get_strategy_prompt_template from langchain.chains import LLMChain from langchain.callbacks import AsyncIteratorCallbackHandler from typing import AsyncIterable, List, Optional import asyncio from langchain.prompts.chat import ChatPromptTemplate from server.chat.utils import History from server.knowledge_base.kb_service.base import KBServiceFactory import json from urllib.parse import urlencode from server.knowledge_base.kb_doc_api import search_docs from server.reranker.reranker import LangchainReranker from server.utils import embedding_device from server.chat.policy_fun import add_summary_retrieved_results, get_llm_model_response import json async def report_chat(query: str = Body(..., description="用户输入", examples=["你好"]), fileName: List = Body([], description="文件名称", examples=[[]]), knowledge_base_name: str = Body(..., description="知识库名称", examples=["t_strategy_report_bge_v1"]), top_k: int = Body(VECTOR_SEARCH_TOP_K, description="匹配向量数"), score_threshold: float = Body( SCORE_THRESHOLD, description="知识库匹配相关度阈值，取值范围在0-1之间，SCORE越小，相关度越高，取到1相当于不筛选，建议设置在0.5左右", ge=0, le=2 ), history: List[History] = Body( [], description="历史对话", examples=[[]] ), stream: bool = Body(False, description="流式输出"), model_name: str = Body(LLM_MODELS[0], description="LLM 模型名称。"), temperature: float = Body(TEMPERATURE, description="LLM 采样温度", ge=0.0, le=1.0), max_tokens: Optional[int] = Body( MAX_TOKENS, description="限制LLM生成Token数量，默认None代表模型最大值" ), prompt_name: str = Body( "default", description="使用的prompt模板名称(在configs/prompt_config.py中配置)" ), request: Request = None, use_summary = True, chunk_size: int = 20000, min_chunk_size: int = 2000, summary_model_name = LLM_MODELS[0], query_rewrite_model_name = LLM_MODELS[0] ): kb = KBServiceFactory.get_service_by_name(knowledge_base_name) if kb is None: return BaseResponse(code=404, msg=f"未找到知识库 {knowledge_base_name}") history = [History.from_data(h) for h in history] async def knowledge_base_chat_iterator( query: str, top_k: int, history: Optional[List[History]], model_name: str = model_name, prompt_name: str = prompt_name, ) -> AsyncIterable[str]: nonlocal max_tokens callback = AsyncIteratorCallbackHandler() if isinstance(max_tokens, int) and max_tokens <= 0: max_tokens = None model = get_ChatOpenAI( model_name=model_name, temperature=temperature, max_tokens=max_tokens, callbacks=[callback], ) # print('-------------- debug', query) search_query = get_llm_model_response( strategy_name="query rewrite", llm_model_name=query_rewrite_model_name, template_prompt_name="query_rewrite_report", prompt_param_dict={"query": query}, temperature=0.01, max_tokens=512 ) # print('search query', search_query) json_string = search_query.strip("```json\n").strip("```") # print('search query----json string', json_string) try: # 防止json格式错误 # 读取改写后的query data = json.loads(json_string) policies = data['report'] search_query = '' for policy in policies: search_query += policy except: search_query = query print('search query', search_query) docs = await run_in_threadpool(search_docs, fileName=fileName, query=search_query, knowledge_base_name=knowledge_base_name, top_k=top_k, score_threshold=score_threshold) # print(docs) # doc加入metadata的summary字段 if use_summary: docs = await add_summary_retrieved_results(docs, query, 512,chunk_size,min_chunk_size,summary_model_name) print(docs) # context = "\n".join([doc.page_content for doc in docs]) # 需要规范格式的prompt_name # 默认default，即为空，不用管 format_list = ["Abstract Assistant", "Outline Assistant"] if prompt_name in format_list: format_template = get_format_template("knowledge_base_chat", "abstract_format") else: format_template = get_format_template("knowledge_base_chat", "default") # 政策知识库 # 相关信息把标题和内容进行整合 if knowledge_base_name == 't_strategy_report_bge_v1': knowledge = [] newdocs =[] for inum,doc in enumerate(docs): if use_summary : if len(doc.metadata['summary'])>15: knowledge.append(f"""参考报告[{len(knowledge) + 1}] 报告来源： {doc.metadata['source']} \n报告内容： {doc.metadata['summary']}""") newdocs.append(doc) else: pass else: knowledge.append(f"""参考报告[{inum + 1}] 报告来源: {doc.metadata['source']} \n报告内容: {doc.page_content}""") context = "\n\n".join(knowledge) docs = newdocs # 非报告知识库 else: context = "\n".join([doc.page_content for doc in docs]) if len(docs) == 0 and fileName: # 如果没有找到相关文档，使用empty模板 prompt_template = get_prompt_template("knowledge_base_chat", prompt_name) elif len(docs) == 0 and not fileName and prompt_name != "Abstract Assistant": prompt_template = get_prompt_template("knowledge_base_chat", "empty") elif prompt_name == 'iast_report_chat' or (knowledge_base_name == "t_strategy_report_bge_v1" and prompt_name == 'default'): print("use report prompt_template") prompt_template = get_strategy_prompt_template("knowledge_base_chat", 'iast_report_chat') else: prompt_template = get_prompt_template("knowledge_base_chat", prompt_name) print("prompt_template", prompt_template) input_msg = History(role="user", content=prompt_template).to_msg_template(False) chat_prompt = ChatPromptTemplate.from_messages( [i.to_msg_template() for i in history] + [input_msg]) chain = LLMChain(prompt=chat_prompt, llm=model) print( f"\n知识库问答开始调用：参数：\nkb:{knowledge_base_name}\nquery:{query}\nhistory:{history}\ncontext:{context}\nfile_name:{fileName}\nformat_template:{format_template}\n\n") query = query.replace("原文", "") task = asyncio.create_task(wrap_done( chain.acall({"context": context, "history": history, "question": query, "file_name": str(fileName), "format_template": format_template}), callback.done), ) source_documents = [] # 报告知识库 if knowledge_base_name == 't_strategy_report_bge_v1': for inum, doc in enumerate(docs): filename = doc.metadata.get("source") print("filename", filename) if filename: text = f"""[{inum + 1}] 报告出处: [{filename}]\n\n{doc.metadata['summary']}\n\n""" else: text = f"""[{inum + 1}] \n\n{doc.metadata['summary']}\n\n""" source_documents.append(text) # 非报告知识库 else: for inum, doc in enumerate(docs): filename = doc.metadata.get("source") parameters = urlencode({"knowledge_base_name": knowledge_base_name, "file_name": filename}) base_url = request.base_url url = f"{base_url}knowledge_base/download_doc?" + parameters if filename: text = f"""出处: [{filename}]({url}) \n\n""" else: text = f"""出处: [{"原文地址"}]({url}) \n\n""" source_documents.append(text) if len(source_documents) == 0: # 没有找到相关文档 source_documents.append(f"未找到相关文档,该回答为大模型自身能力解答！") if stream: async for token in callback.aiter(): # Use server-sent-events to stream the response yield json.dumps({"answer": token}, ensure_ascii=False) else: answer = "" async for token in callback.aiter(): answer += token yield json.dumps({"answer": answer}) await task yield json.dumps({"docs": source_documents}, ensure_ascii=False) return EventSourceResponse(knowledge_base_chat_iterator(query, top_k, history, model_name, prompt_name))