2023-04-13 23:01:52 +08:00
|
|
|
from configs.model_config import *
|
|
|
|
|
from chains.local_doc_qa import LocalDocQA
|
2023-04-16 23:38:25 +08:00
|
|
|
import os
|
|
|
|
|
import nltk
|
|
|
|
|
|
|
|
|
|
nltk.data.path = [os.path.join(os.path.dirname(__file__), "nltk_data")] + nltk.data.path
|
2023-04-13 23:01:52 +08:00
|
|
|
|
|
|
|
|
# return top-k text chunk from vector store
|
2023-04-17 23:59:22 +08:00
|
|
|
VECTOR_SEARCH_TOP_K = 6
|
2023-04-13 23:01:52 +08:00
|
|
|
|
|
|
|
|
# LLM input history length
|
|
|
|
|
LLM_HISTORY_LEN = 3
|
|
|
|
|
|
|
|
|
|
# Show reply with source text from input document
|
|
|
|
|
REPLY_WITH_SOURCE = True
|
|
|
|
|
|
|
|
|
|
if __name__ == "__main__":
|
|
|
|
|
local_doc_qa = LocalDocQA()
|
|
|
|
|
local_doc_qa.init_cfg(llm_model=LLM_MODEL,
|
|
|
|
|
embedding_model=EMBEDDING_MODEL,
|
|
|
|
|
embedding_device=EMBEDDING_DEVICE,
|
2023-04-14 00:06:45 +08:00
|
|
|
llm_history_len=LLM_HISTORY_LEN,
|
|
|
|
|
top_k=VECTOR_SEARCH_TOP_K)
|
2023-04-13 23:01:52 +08:00
|
|
|
vs_path = None
|
|
|
|
|
while not vs_path:
|
|
|
|
|
filepath = input("Input your local knowledge file path 请输入本地知识文件路径：")
|
|
|
|
|
vs_path = local_doc_qa.init_knowledge_vector_store(filepath)
|
|
|
|
|
history = []
|
|
|
|
|
while True:
|
|
|
|
|
query = input("Input your question 请输入问题：")
|
|
|
|
|
resp, history = local_doc_qa.get_knowledge_based_answer(query=query,
|
|
|
|
|
vs_path=vs_path,
|
|
|
|
|
chat_history=history)
|
|
|
|
|
if REPLY_WITH_SOURCE:
|
|
|
|
|
print(resp)
|
|
|
|
|
else:
|
|
|
|
|
print(resp["result"])
|
