forked from phito/darknet_diaries_llm
107 lines
3.6 KiB
Python
107 lines
3.6 KiB
Python
from llama_index import (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, ChatMessage, MessageRole
|
|
from llama_index.prompts import ChatPromptTemplate
|
|
from llama_index import set_global_handler
|
|
from llama_index.chat_engine.types import ChatMode
|
|
import os
|
|
import re
|
|
|
|
#set_global_handler("simple")
|
|
|
|
llm = OpenAI(model="gpt-4", temperature=0, max_tokens=256)
|
|
service_context = ServiceContext.from_defaults(llm=llm)
|
|
set_global_service_context(service_context)
|
|
|
|
if not os.path.exists("./index/lock"):
|
|
documents = []
|
|
for filename in os.listdir("./transcripts"):
|
|
episode_number = re.search(r'\d+', filename).group()
|
|
with open("./transcripts/" + filename, 'r') as f:
|
|
title = f.readline().strip()
|
|
downloads = f.readline().strip()
|
|
content = f.read()
|
|
document = Document(
|
|
text=content,
|
|
doc_id=filename,
|
|
metadata={
|
|
"episode_number": episode_number,
|
|
"episode_title": title,
|
|
"episode_downloads": downloads,
|
|
"episode_url": f"https://darknetdiaries.com/episode/{episode_number}/"
|
|
}
|
|
)
|
|
documents.append(document)
|
|
|
|
parser = SimpleNodeParser.from_defaults()
|
|
nodes = parser.get_nodes_from_documents(documents)
|
|
|
|
index = VectorStoreIndex(nodes, show_progress=True)
|
|
index.storage_context.persist(persist_dir="./index")
|
|
open("./index/lock", 'a').close()
|
|
else:
|
|
print("Loading index...")
|
|
storage_context = StorageContext.from_defaults(persist_dir="./index")
|
|
index = load_index_from_storage(storage_context)
|
|
|
|
chat_text_qa_msgs = [
|
|
ChatMessage(
|
|
role=MessageRole.SYSTEM,
|
|
content=(
|
|
"You have been trained on the Darknet Diaries podcast transcripts with data from october 6 2023."
|
|
"You are an expert about it and will answer as such. You know about every episode up to number 138."
|
|
"Always answer the question, even if the context isn't helpful."
|
|
"Mention the number and title of the episodes you are referring to."
|
|
)
|
|
),
|
|
ChatMessage(
|
|
role=MessageRole.USER,
|
|
content=(
|
|
"Context information is below.\n"
|
|
"---------------------\n"
|
|
"{context_str}\n"
|
|
"---------------------\n"
|
|
"Given the context information and not prior knowledge,"
|
|
"answer the question: {query_str}\n"
|
|
)
|
|
)
|
|
]
|
|
text_qa_template = ChatPromptTemplate(chat_text_qa_msgs)
|
|
|
|
chat_refine_msgs = [
|
|
ChatMessage(
|
|
role=MessageRole.SYSTEM,
|
|
content="Always answer the question, even if the context isn't helpful.",
|
|
),
|
|
ChatMessage(
|
|
role=MessageRole.USER,
|
|
content=(
|
|
"We have the opportunity to refine the original answer "
|
|
"(only if needed) with some more context below.\n"
|
|
"------------\n"
|
|
"{context_msg}\n"
|
|
"------------\n"
|
|
"Given the new context, refine the original answer to better "
|
|
"answer the question: {query_str}. "
|
|
"If the context isn't useful, output the original answer again.\n"
|
|
"Original Answer: {existing_answer}"
|
|
),
|
|
),
|
|
]
|
|
refine_template = ChatPromptTemplate(chat_refine_msgs)
|
|
|
|
chat_engine = index.as_chat_engine(
|
|
text_qa_template=text_qa_template,
|
|
refine_template=refine_template,
|
|
chat_mode=ChatMode.OPENAI
|
|
)
|
|
|
|
while True:
|
|
try:
|
|
chat_engine.chat_repl()
|
|
except KeyboardInterrupt:
|
|
break
|
|
|