forked from phito/darknet_diaries_llm
Chat mode
This commit is contained in:
parent
b41d8288b7
commit
3d734a3064
28
main.py
28
main.py
@ -2,7 +2,9 @@ from llama_index import (SimpleDirectoryReader, ServiceContext, StorageContext,
|
||||
load_index_from_storage, Document, set_global_service_context)
|
||||
from llama_index.node_parser import SimpleNodeParser
|
||||
from llama_index import VectorStoreIndex
|
||||
from llama_index.llms import OpenAI
|
||||
from llama_index.llms import OpenAI, ChatMessage, MessageRole
|
||||
from llama_index.prompts import PromptTemplate
|
||||
from llama_index.chat_engine.condense_question import CondenseQuestionChatEngine
|
||||
import os
|
||||
import re
|
||||
|
||||
@ -36,22 +38,32 @@ else:
|
||||
storage_context = StorageContext.from_defaults(persist_dir="./index")
|
||||
index = load_index_from_storage(storage_context)
|
||||
|
||||
template = (
|
||||
custom_prompt = PromptTemplate(
|
||||
"You have been trained on the Darknet Diaries podcast transcripts with data from october 6 2023."
|
||||
"You are now an expert about it and will answer as such. You know about every episode up to number 138. \n"
|
||||
"----------------\n"
|
||||
"Here is the context: {context_str}"
|
||||
"Chat history: {chat_history}\n"
|
||||
"----------------\n"
|
||||
"Please answer this question by referring to the podcast: {query_str}"
|
||||
"Please answer this question by referring to the podcast: {question}"
|
||||
)
|
||||
qa_template = PromptTemplate(template)
|
||||
query_engine = index.as_query_engine(text_qa_template=qa_template)
|
||||
|
||||
custom_chat_history = []
|
||||
query_engine = index.as_query_engine()
|
||||
chat_engine = CondenseQuestionChatEngine.from_defaults(
|
||||
query_engine=query_engine,
|
||||
condense_question_prompt=custom_prompt,
|
||||
chat_history=custom_chat_history,
|
||||
verbose=True
|
||||
)
|
||||
|
||||
|
||||
while True:
|
||||
try:
|
||||
user_prompt = input("Prompt: ")
|
||||
response = query_engine.query(user_prompt)
|
||||
print(response)
|
||||
streaming_response = chat_engine.stream_chat(user_prompt)
|
||||
for token in streaming_response.response_gen:
|
||||
print(token, end="")
|
||||
print("\n")
|
||||
except KeyboardInterrupt:
|
||||
break
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user