565 lines
34 KiB
Python
565 lines
34 KiB
Python
from fastapi import Body, Request
|
||
from langchain.chains.question_answering import load_qa_chain
|
||
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,
|
||
POLICY_KNOWLEDGE_BASE,
|
||
REPORT_KNOWLEDGE_BASE,
|
||
JOURNAL_KNOWLEDGE_BASE,
|
||
OLD_POLICY_BASE
|
||
)
|
||
from configs.kb_config import OLD_JOURNAL_BASE
|
||
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
|
||
from server.chat.policy_fun_iast import get_llm_model_response
|
||
import json
|
||
from langchain.memory import ConversationSummaryBufferMemory, ConversationBufferWindowMemory, ConversationBufferMemory
|
||
from langchain_core.prompts import PromptTemplate
|
||
import itertools
|
||
from datetime import datetime
|
||
import time
|
||
from langchain.schema import Document
|
||
|
||
REPLACEMENT_RULES = [
|
||
(OLD_POLICY_BASE, "t_policy_total_bge_new_v2"),
|
||
(OLD_JOURNAL_BASE, "t_journal_article_bge_v1")
|
||
]
|
||
async def knowledge_base_chat(query: str = Body(..., description="用户输入", examples=["你好"]),
|
||
fileName: List = Body([], description="文件名称", examples=[["123.txt"]]),
|
||
knowledge_base_name_list: list = Body(..., description="多种知识库名称",
|
||
examples=[[ "t_policy_total_bge_v1","t_strategy_report_20_bge_v2","t_journal_article_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=[[
|
||
{"role": "user",
|
||
"content": "我们来玩成语接龙，我先来，生龙活虎"},
|
||
{"role": "assistant",
|
||
"content": "虎头虎脑"}]]
|
||
),
|
||
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,
|
||
use_model_self_response = True,
|
||
chunk_size: int = 20000,
|
||
min_chunk_size: int = 2000,
|
||
summary_model_name = LLM_MODELS[0],
|
||
query_rewrite_model_name = LLM_MODELS[0]
|
||
):
|
||
# 创建集合提高查找效率
|
||
original_kb_set = set(knowledge_base_name_list)
|
||
new_elements_added = []
|
||
|
||
# 批量处理替换规则
|
||
for old_bases, new_base in REPLACEMENT_RULES:
|
||
# 使用集合运算快速找到需要移除的元素
|
||
to_remove = original_kb_set & set(old_bases)
|
||
|
||
if to_remove:
|
||
# 使用列表推导式生成新列表（保持原有顺序）
|
||
knowledge_base_name_list = [
|
||
elem for elem in knowledge_base_name_list
|
||
if elem not in to_remove
|
||
]
|
||
new_elements_added.append(new_base)
|
||
|
||
# 去重后添加新元素（如果原列表已存在则不添加）
|
||
for new_base in new_elements_added:
|
||
if new_base not in knowledge_base_name_list:
|
||
knowledge_base_name_list.append(new_base)
|
||
print(f'========== 当前检索的知识库：{knowledge_base_name_list} ==========')
|
||
new_knowledge_base_name_list = knowledge_base_name_list[:]
|
||
for knowledge_base_name in knowledge_base_name_list:
|
||
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]
|
||
# 记录开始时间
|
||
start_time = time.time()
|
||
history = [History.from_data(h) for h in history]
|
||
print(f"========== 当前的对话历史为==========\n{history}")
|
||
# 获取当前时间并格式化为YYYYMMDD
|
||
current_time = datetime.now().strftime("%Y%m%d")
|
||
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()
|
||
memory = None
|
||
policydocs = []
|
||
reportdocs = []
|
||
journaldocs = []
|
||
personaldocs = []
|
||
docs = []
|
||
if isinstance(max_tokens, int) and max_tokens <= 0:
|
||
max_tokens = None
|
||
if prompt_name == "policy_chat":
|
||
model_name = LLM_MODELS[0]
|
||
model = get_ChatOpenAI(
|
||
model_name=model_name,
|
||
temperature=temperature,
|
||
max_tokens=max_tokens,
|
||
callbacks=[callback],
|
||
)
|
||
knowledge = []
|
||
self_knowledge = []
|
||
user_queries = [] # 初始化列表来收集用户消息
|
||
if use_model_self_response:
|
||
# 获取大模型本身对用户问题的回答
|
||
modelself_response=get_llm_model_response(
|
||
strategy_name="self response",
|
||
llm_model_name=query_rewrite_model_name,
|
||
template_prompt_name="self_response",
|
||
prompt_param_dict={"query": query},
|
||
temperature=0.01,
|
||
max_tokens=512
|
||
)
|
||
self_knowledge.append(f"""{modelself_response}""")
|
||
if len(knowledge_base_name_list) != 0:
|
||
# 政策库
|
||
if POLICY_KNOWLEDGE_BASE in knowledge_base_name_list:
|
||
# 遍历历史消息并收集用户消息
|
||
for message in history:
|
||
if message.role == 'user':
|
||
user_queries.append(message.content)
|
||
#改写原问题
|
||
search_query = get_llm_model_response(
|
||
strategy_name="query rewrite",
|
||
llm_model_name=query_rewrite_model_name,
|
||
template_prompt_name="query_rewrite_policy",
|
||
prompt_param_dict={"query": query, "history": user_queries, "time": current_time},
|
||
temperature=0.01,
|
||
max_tokens=512
|
||
)
|
||
print("search_query: ", query)
|
||
print("search_history: ", user_queries)
|
||
json_string = search_query.strip("```json\n").strip("```")
|
||
try: # 防止json格式错误
|
||
# 读取改写后的query
|
||
data = json.loads(json_string)
|
||
policies = data['policies']
|
||
search_query = ''
|
||
for policy in policies:
|
||
search_query += policy
|
||
except:
|
||
search_query = query
|
||
|
||
print('policy search query', search_query)
|
||
#搜索政策相关的docs
|
||
policydocs = await run_in_threadpool(search_docs,
|
||
fileName=fileName,
|
||
query=search_query,
|
||
usr_query=query,
|
||
knowledge_base_name=POLICY_KNOWLEDGE_BASE,
|
||
top_k=top_k,
|
||
score_threshold=score_threshold)
|
||
# print('政策数据库共搜索出：',len(policydocs))
|
||
#使用概括将只有文章标题的内容总结成段落
|
||
if use_summary:
|
||
# policydocs = await add_summary_retrieved_results(policydocs, query, 512,chunk_size,min_chunk_size,summary_model_name)
|
||
seen_docs = set() # 用于跟踪已见过的标题和内容组合
|
||
duplicate_indices = [] # 用于跟踪重复文档的索引
|
||
for inum,doc in enumerate(policydocs):
|
||
if len(doc.metadata['summary'])>15:
|
||
doc_identifier = (doc.metadata['title'], doc.page_content)
|
||
# 检查此标识符是否已存在于集合中
|
||
if doc_identifier not in seen_docs:
|
||
# 如果不存在，将其添加到集合中
|
||
seen_docs.add(doc_identifier)
|
||
knowledge.append(f"""参考资料[{len(knowledge) + 1}] 文章标题： {doc.metadata['title']} \n文章内容： {doc.metadata['summary']}""")
|
||
else:
|
||
# 如果存在，将当前索引添加到重复索引列表中
|
||
duplicate_indices.append(inum)
|
||
else:
|
||
duplicate_indices.append(inum)
|
||
# 从policydocs中删除重复的文档（从后往前删除以防止索引错位）
|
||
for index in sorted(duplicate_indices, reverse=True):
|
||
del policydocs[index]
|
||
else:
|
||
for inum,doc in enumerate(policydocs):
|
||
if doc.metadata["_type"] == "title":
|
||
knowledge.append(f"""参考资料[{inum + 1}] 文章标题 {doc.page_content} \n文章内容 {doc.metadata['content']}""")
|
||
if doc.metadata["_type"] == "content":
|
||
knowledge.append(f"""参考资料[{inum + 1}] 文章标题 {doc.metadata['title']} \n文章内容 {doc.page_content}""")
|
||
new_knowledge_base_name_list.remove(POLICY_KNOWLEDGE_BASE)
|
||
# print('政策数据库剩下：',len(policydocs))
|
||
# 报告库
|
||
if REPORT_KNOWLEDGE_BASE in knowledge_base_name_list:
|
||
# 遍历历史消息并收集用户消息
|
||
for message in history:
|
||
if message.role == 'user':
|
||
user_queries.append(message.content)
|
||
#先改写原问题
|
||
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, "history": user_queries + [query]},
|
||
temperature=0.01,
|
||
max_tokens=512
|
||
)
|
||
print("search_query: ", query)
|
||
print("search_history: ", user_queries)
|
||
json_string = search_query.strip("```json\n").strip("```")
|
||
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('report search query', search_query)
|
||
reportdocs = await run_in_threadpool(search_docs,
|
||
fileName=fileName,
|
||
query=search_query,
|
||
knowledge_base_name=REPORT_KNOWLEDGE_BASE,
|
||
top_k=top_k,
|
||
score_threshold=score_threshold,
|
||
expr = " _type == 'content'")
|
||
# print('报告数据库共搜索出：',len(reportdocs))
|
||
seen_docs = set() # 用于跟踪已见过的标题和内容组合
|
||
duplicate_indices = [] # 用于跟踪重复文档的索引
|
||
for inum,doc in enumerate(reportdocs):
|
||
doc_identifier = (doc.metadata['source'], doc.page_content)
|
||
# 检查此标识符是否已存在于集合中
|
||
if doc_identifier not in seen_docs:
|
||
# 如果不存在，将其添加到集合中
|
||
seen_docs.add(doc_identifier)
|
||
# 并将文档信息添加到knowledge列表中
|
||
knowledge.append(f"""参考资料[{len(knowledge) + 1}] 报告来源： {doc.metadata['source'].replace('.pdf','')} \n报告内容： {doc.page_content}""")
|
||
else:
|
||
duplicate_indices.append(inum)
|
||
# print('重复报告',doc_identifier)
|
||
# 从reportdocs中删除重复的文档（从后往前删除以防止索引错位）
|
||
for index in sorted(duplicate_indices, reverse=True):
|
||
del reportdocs[index]
|
||
new_knowledge_base_name_list.remove(REPORT_KNOWLEDGE_BASE)
|
||
# 期刊库
|
||
if JOURNAL_KNOWLEDGE_BASE in knowledge_base_name_list:
|
||
# 遍历历史消息并收集用户消息
|
||
for message in history:
|
||
if message.role == 'user':
|
||
user_queries.append(message.content)
|
||
#先改写原问题
|
||
search_query = get_llm_model_response(
|
||
strategy_name="query rewrite",
|
||
llm_model_name=query_rewrite_model_name,
|
||
template_prompt_name="query_rewrite",
|
||
prompt_param_dict={"query": query, "history": user_queries + [query]},
|
||
temperature=0.01,
|
||
max_tokens=512
|
||
)
|
||
print("search_query: ", query)
|
||
print("search_history: ", user_queries)
|
||
json_string = search_query.strip("```json\n").strip("```")
|
||
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('journal search query', search_query)
|
||
journaldocs = await run_in_threadpool(search_docs,
|
||
fileName=fileName,
|
||
query=search_query,
|
||
knowledge_base_name=JOURNAL_KNOWLEDGE_BASE,
|
||
top_k=top_k,
|
||
score_threshold=score_threshold)
|
||
# print('期刊数据库共搜索出：',len(journaldocs))
|
||
seen_docs = set() # 用于跟踪已见过的标题和内容组合
|
||
duplicate_indices = [] # 用于跟踪重复文档的索引
|
||
for inum,doc in enumerate(journaldocs):
|
||
doc_identifier = (doc.metadata['title'], doc.metadata['abstract'])
|
||
# 检查此标识符是否已存在于集合中
|
||
if doc_identifier not in seen_docs:
|
||
# 如果不存在，将其添加到集合中
|
||
seen_docs.add(doc_identifier)
|
||
# 并将文档信息添加到knowledge列表中
|
||
knowledge.append(f"""参考资料[{len(knowledge) + 1}] 论文标题： {doc.metadata['title']} \n论文摘要： {doc.metadata['abstract']}""")
|
||
else:
|
||
duplicate_indices.append(inum)
|
||
# print('重复期刊',doc_identifier)
|
||
# 从journaldocs中删除重复的文档（从后往前删除以防止索引错位）
|
||
for index in sorted(duplicate_indices, reverse=True):
|
||
del journaldocs[index]
|
||
new_knowledge_base_name_list.remove(JOURNAL_KNOWLEDGE_BASE)
|
||
if len(new_knowledge_base_name_list)>0:
|
||
# 个人知识库
|
||
for knowledge_base_name in new_knowledge_base_name_list:
|
||
if knowledge_base_name == 'yj_oa_journal_bge_v2_yejinbak':
|
||
knowledge_base_name = 'yj_oa_article_v1_yejinbak' #采集数据代替oa资源
|
||
personaldocs = await run_in_threadpool(search_docs,
|
||
fileName=fileName,
|
||
query=query,
|
||
knowledge_base_name=knowledge_base_name,
|
||
top_k=top_k,
|
||
score_threshold=score_threshold)
|
||
seen_docs = set() # 用于跟踪已见过的标题和内容组合
|
||
for inum,doc in enumerate(personaldocs):
|
||
doc_identifier = (doc.page_content)
|
||
# 检查此标识符是否已存在于集合中
|
||
if doc_identifier not in seen_docs:
|
||
# 如果不存在，将其添加到集合中
|
||
seen_docs.add(doc_identifier)
|
||
# 并将文档信息添加到knowledge列表中
|
||
knowledge.append(f"""参考资料[{len(knowledge) + 1}] {doc.page_content}""")
|
||
else:
|
||
personaldocs = await run_in_threadpool(search_docs,
|
||
fileName=fileName,
|
||
query=query,
|
||
knowledge_base_name=knowledge_base_name,
|
||
top_k=top_k,
|
||
score_threshold=score_threshold)
|
||
seen_docs = set() # 用于跟踪已见过的标题和内容组合
|
||
for inum,doc in enumerate(personaldocs):
|
||
doc_identifier = (doc.page_content)
|
||
# 检查此标识符是否已存在于集合中
|
||
if doc_identifier not in seen_docs:
|
||
# 如果不存在，将其添加到集合中
|
||
seen_docs.add(doc_identifier)
|
||
# 并将文档信息添加到knowledge列表中
|
||
knowledge.append(f"""参考资料[{len(knowledge) + 1}] {doc.page_content}""")
|
||
# context = "\n\n".join(knowledge)
|
||
docs = [Document(page_content=k) for k in knowledge]
|
||
# print(f"=========================知识库问答参考资料====================\n{docs}\n====================知识库问答参考资料====================")
|
||
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 len(knowledge) == 0 and not fileName and prompt_name != "Abstract Assistant":
|
||
prompt_template = get_prompt_template("knowledge_base_chat", "empty")
|
||
# elif prompt_name == 'default' and "t_policy_total_bge_v1" in knowledge_base_name_list:
|
||
# if len(knowledge_base_name_list) == 1: # 如果是科学研究院policy推荐功能，则使用如下模板
|
||
# prompt_template = get_strategy_prompt_template("knowledge_base_chat", 'iast_policy_chat')
|
||
# else:
|
||
# prompt_template = get_prompt_template("knowledge_base_chat", prompt_name)
|
||
else:
|
||
prompt_template = get_prompt_template("knowledge_base_chat", prompt_name)
|
||
print("prompt_name(no history):", prompt_name)
|
||
if history and prompt_name not in ["Question Assistant"]:
|
||
prompt_template = get_prompt_template("knowledge_base_chat", prompt_name)
|
||
print("prompt_name(with history):", prompt_name)
|
||
chat_prompt = PromptTemplate.from_template(template=prompt_template, template_format='jinja2')
|
||
# 把history转成memory
|
||
memory = ConversationBufferMemory(memory_key="history", input_key="question")
|
||
for message in history:
|
||
# 检查消息的角色
|
||
if message.role == 'user':
|
||
# 添加用户消息
|
||
memory.chat_memory.add_user_message(message.content)
|
||
elif message.role == 'assistant':
|
||
# 添加AI消息
|
||
memory.chat_memory.add_ai_message(message.content)
|
||
else:
|
||
input_prompt = History(role="system", content=prompt_template).to_msg_template(False)
|
||
chat_prompt = ChatPromptTemplate.from_messages([input_prompt])
|
||
|
||
query = query.replace("原文", "")
|
||
chain = load_qa_chain(
|
||
model, chain_type="stuff", memory=memory, prompt=chat_prompt, verbose=True
|
||
)
|
||
# docs = list(itertools.chain(policydocs, reportdocs, journaldocs, personaldocs))
|
||
task = asyncio.create_task(wrap_done(
|
||
chain.acall({
|
||
# "context": context,
|
||
"input_documents": docs,
|
||
"self_knowledge":self_knowledge,
|
||
"history": history,
|
||
"question": query,
|
||
"file_name": str(fileName),
|
||
"format_template": format_template,
|
||
"time": current_time
|
||
}),
|
||
callback.done),
|
||
)
|
||
source_documents = []
|
||
|
||
if len(knowledge_base_name_list) != 0:
|
||
# 政策库
|
||
if POLICY_KNOWLEDGE_BASE in knowledge_base_name_list:
|
||
for inum, doc in enumerate(policydocs):
|
||
# 获取标题以及详情地址（url）
|
||
filename = doc.metadata.get("title")
|
||
# detail_url = doc.metadata.get("source")
|
||
detail_url = "https://policy.ckcest.cn/detail/" + doc.metadata.get("primary_key") + ".html"
|
||
# 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
|
||
# text = f"""出处 [{inum + 1}] [{filename}]({url}) \n\n{doc.page_content}\n\n"""
|
||
if filename:
|
||
# print(doc.metadata.get('_type'), detail_url)
|
||
# if doc.metadata.get('_type') == 'title':
|
||
filename = filename.replace('\r', '').replace('\n', '')
|
||
text = f"""_政策[{len(source_documents) + 1}] [{filename}]({detail_url})_\n"""
|
||
# else:
|
||
# text = f"""政策: [{len(source_documents) + 1}][{filename}]({detail_url})\n\n{doc.page_content} \n\n"""
|
||
else:
|
||
# if doc.metadata.get('_type') == 'title':
|
||
text = f"""_政策[{len(source_documents) + 1}] [{"原文地址"}]({detail_url})_"""
|
||
# else:
|
||
# text = f"""政策: [{len(source_documents) + 1}][{"原文地址"}]({detail_url})\n\n{doc.page_content}\n\n"""
|
||
source_documents.append(text)
|
||
# 报告库
|
||
if REPORT_KNOWLEDGE_BASE in knowledge_base_name_list:
|
||
for inum, doc in enumerate(reportdocs):
|
||
text = f"""_报告[{len(source_documents) + 1}] [{doc.metadata.get("source").replace('.pdf','')}](https://kgo.ckcest.cn/kgo/list?dbId=1010&word=&shortName=ALL&page=1&order=1)_"""
|
||
source_documents.append(text)
|
||
# 期刊库
|
||
if JOURNAL_KNOWLEDGE_BASE in knowledge_base_name_list:
|
||
for inum, doc in enumerate(journaldocs):
|
||
text = f"""_期刊论文[{len(source_documents) + 1}] [{doc.metadata.get("title")}](https://kgo.ckcest.cn/kgo/detail/1002/ads_journal_article/{doc.metadata.get("ID")}.html)_"""
|
||
source_documents.append(text)
|
||
|
||
|
||
#个人知识库
|
||
if len(new_knowledge_base_name_list)>0:
|
||
for knowledge_base_name in new_knowledge_base_name_list:
|
||
if knowledge_base_name == 'yj_oa_journal_bge_v2_yejinbak':
|
||
knowledge_base_name = 'yj_oa_article_v1_yejinbak' #采集数据代替oa资源
|
||
docs = await run_in_threadpool(search_docs,
|
||
fileName=fileName,
|
||
query=query,
|
||
knowledge_base_name=knowledge_base_name,
|
||
top_k=top_k,
|
||
score_threshold=score_threshold)
|
||
|
||
seen_docs = set() # 用于跟踪已见过的内容组合
|
||
for inum,doc in enumerate(docs):
|
||
doc_identifier = (doc.page_content)
|
||
hasSummary = doc.metadata.get("summary")
|
||
if doc_identifier not in seen_docs:
|
||
# 如果不存在，将其添加到集合中
|
||
seen_docs.add(doc_identifier)
|
||
if doc.metadata.get('_type') == 'title' and hasSummary and knowledge_base_name in ["yj_policys_bge_v1_yejinbak","yj_oa_journal_bge_v2_yejinbak","yj_for_journal_bge_v1_yejinbak","yj_ch_journal_bge_v1_yejinbak"]:
|
||
text = f"""[{len(source_documents) + 1}] 《{doc.page_content}》\n{doc.metadata.get("summary")}\n资料年份：{doc.metadata.get("publish_year")}\n\n"""
|
||
elif doc.metadata.get('_type') == 'title' and knowledge_base_name in ["yj_policys_bge_v1_yejinbak","yj_oa_journal_bge_v2_yejinbak","yj_for_journal_bge_v1_yejinbak","yj_ch_journal_bge_v1_yejinbak"]:
|
||
text = f"""[{len(source_documents) + 1}] 《{doc.page_content}》\n资料年份：{doc.metadata.get("publish_year")}\n\n"""
|
||
elif knowledge_base_name in ["yj_policys_bge_v1_yejinbak","yj_oa_journal_bge_v2_yejinbak","yj_for_journal_bge_v1_yejinbak","yj_ch_journal_bge_v1_yejinbak"]:
|
||
text = f"""[{len(source_documents) + 1}] 《{doc.metadata.get("title")}》\n资料年份：{doc.metadata.get("publish_year")}\n\n"""
|
||
else:
|
||
# text = f"""参考文档[{len(source_documents) + 1}]： 《{doc.metadata.get("source", "").split('.')[0]}》"""
|
||
text = f"""参考文档[{len(source_documents) + 1}] [{doc.metadata.get("source")}]()\n"""
|
||
source_documents.append(text)
|
||
else:
|
||
docs = await run_in_threadpool(search_docs,
|
||
fileName=fileName,
|
||
query=query,
|
||
knowledge_base_name=knowledge_base_name,
|
||
top_k=top_k,
|
||
score_threshold=score_threshold)
|
||
|
||
seen_docs = set() # 用于跟踪已见过的内容组合
|
||
for inum,doc in enumerate(docs):
|
||
doc_identifier = (doc.page_content)
|
||
hasSummary = doc.metadata.get("summary")
|
||
if doc_identifier not in seen_docs:
|
||
# 如果不存在，将其添加到集合中
|
||
seen_docs.add(doc_identifier)
|
||
if doc.metadata.get('_type') == 'title' and hasSummary and knowledge_base_name in ["yj_policys_bge_v1_yejinbak","yj_oa_journal_bge_v2_yejinbak","yj_for_journal_bge_v1_yejinbak","yj_ch_journal_bge_v1_yejinbak"]:
|
||
text = f"""[{len(source_documents) + 1}] 《{doc.page_content}》\n{doc.metadata.get("summary")}\n资料年份：{doc.metadata.get("publish_year")}\n\n"""
|
||
elif doc.metadata.get('_type') == 'title' and knowledge_base_name in ["yj_policys_bge_v1_yejinbak","yj_oa_journal_bge_v2_yejinbak","yj_for_journal_bge_v1_yejinbak","yj_ch_journal_bge_v1_yejinbak"]:
|
||
text = f"""[{len(source_documents) + 1}] 《{doc.page_content}》\n资料年份：{doc.metadata.get("publish_year")}\n\n"""
|
||
elif knowledge_base_name in ["yj_policys_bge_v1_yejinbak","yj_oa_journal_bge_v2_yejinbak","yj_for_journal_bge_v1_yejinbak","yj_ch_journal_bge_v1_yejinbak"]:
|
||
text = f"""[{len(source_documents) + 1}] 《{doc.metadata.get("title")}》\n资料年份：{doc.metadata.get("publish_year")}\n\n"""
|
||
else:
|
||
# text = f"""参考文档[{len(source_documents) + 1}]： 《{doc.metadata.get("source", "").split('.')[0]}》"""
|
||
text = f"""参考文档[{len(source_documents) + 1}] [{doc.metadata.get("source")}]()\n"""
|
||
source_documents.append(text)
|
||
# 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"<span style='color:red'>未找到相关文档,该回答为大模型自身能力解答！</span>")
|
||
|
||
first_token = True # 记录是否为第一个token
|
||
if stream:
|
||
answer = ""
|
||
async for token in callback.aiter():
|
||
if first_token:
|
||
first_token = False
|
||
# 记录第一个token返回的时间
|
||
time_elapsed = time.time() - start_time
|
||
print(f"接收响应到模型吐出第一个字耗时: {time_elapsed:.2f} seconds")
|
||
# Use server-sent-events to stream the response
|
||
answer += token
|
||
yield json.dumps({"answer": token}, ensure_ascii=False)
|
||
# print(f'====返回结果====\n {answer}')
|
||
print(f'=====知识库问答模型返回结果=====\n {answer}')
|
||
else:
|
||
answer = ""
|
||
async for token in callback.aiter():
|
||
if first_token:
|
||
first_token = False
|
||
# 记录第一个token返回的时间
|
||
time_elapsed = time.time() - start_time
|
||
print(f"接收响应到模型吐出第一个字耗时: {time_elapsed:.2f} seconds")
|
||
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))
|