88 lines
2.5 KiB
Python
88 lines
2.5 KiB
Python
import json
|
|
import proto.documents_pb2
|
|
from langchain import text_splitter
|
|
from langchain.chains.qa_with_sources import load_qa_with_sources_chain
|
|
from langchain.embeddings import OpenAIEmbeddings
|
|
from langchain.llms.openai import OpenAI
|
|
from langchain.schema.document import Document
|
|
from pymilvus import (
|
|
connections,
|
|
utility,
|
|
FieldSchema,
|
|
CollectionSchema,
|
|
DataType,
|
|
Collection,
|
|
)
|
|
|
|
import init
|
|
import doc_client
|
|
|
|
#
|
|
# question = """
|
|
# yarn : File C:\\Users\\ivamp\\AppData\\Roaming\\npm\\yarn.ps1 cannot be loaded because running scripts is disabled on this sy
|
|
# stem. For more information, see about_Execution_Policies at https:/go.microsoft.com/fwlink/?LinkID=135170.
|
|
# At line:1 char:1
|
|
# + yarn config set registry https://registry.npm.taobao.org/
|
|
# + ~~~~
|
|
# + CategoryInfo : SecurityError: (:) [], PSSecurityException
|
|
# + FullyQualifiedErrorId : UnauthorizedAccess
|
|
#
|
|
# 是什么问题，该怎么解决
|
|
# """
|
|
|
|
question = """
|
|
为什么我会在 WHMCS 下开发摸不着头脑
|
|
"""
|
|
|
|
vec = init.text_to_vector(question)
|
|
|
|
# vec = ""
|
|
#
|
|
# with open("../question_vec.json", "r") as f:
|
|
# vec = json.load(f)
|
|
|
|
search_param = {
|
|
"data": [vec],
|
|
"anns_field": "vector",
|
|
"param": {"metric_type": "L2"},
|
|
"limit": 10,
|
|
"expr": "user_id == 2",
|
|
"output_fields": ["todo_id", "title", "source", "todo_description", "language", "text", "user_id"],
|
|
}
|
|
res = init.collection.search(**search_param)
|
|
|
|
document_chunk_ids = []
|
|
real_document = []
|
|
|
|
for i in range(len(res[0])):
|
|
_chunk_id = res[0][i].id
|
|
print("正在获取分块 " + str(_chunk_id) + " 的内容...")
|
|
|
|
try:
|
|
_chunk_content = doc_client.stub.GetDocumentChunk(proto.documents_pb2.GetDocumentChunkByIdRequest(
|
|
id=_chunk_id
|
|
))
|
|
|
|
# print(_chunk_content)
|
|
|
|
_doc_content_full = _chunk_content.content
|
|
|
|
# real_document.append(_doc_content)
|
|
# doc_obj = Document(page_content=_doc_content_full, metadata={"source": _chunk_content.title})
|
|
doc_obj = Document(page_content=_doc_content_full, metadata={"source": "chunked content"})
|
|
|
|
real_document.append(doc_obj)
|
|
|
|
except Exception as e:
|
|
print(e)
|
|
|
|
print(real_document)
|
|
|
|
print("正在调用 LLM...")
|
|
chain = load_qa_with_sources_chain(OpenAI(temperature=0), chain_type="map_reduce", return_intermediate_steps=True,
|
|
verbose=True)
|
|
|
|
question = "必须使用中文回复:" + question
|
|
output = chain({"input_documents": real_document, "question": question}, return_only_outputs=False)
|
|
print("回复：" + output["output_text"])
|