llm/utils.py

from langchain_community.llms import HuggingFaceEndpoint
from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationalRetrievalChain
import gradio as gr
import os
from llm.CustomRetriever import CustomRetriever

from langchain.schema.retriever import BaseRetriever
from langchain_core.documents import Document
from typing import List

from langchain.callbacks.manager import CallbackManagerForRetrieverRun

from langchain_core.documents import Document
from langchain_core.runnables import chain

API_TOKEN=os.getenv("TOKEN")


# Initialize langchain LLM chain
def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vdb, 
                        thold=0.8, progress=gr.Progress()):
    
    llm = HuggingFaceEndpoint(
            huggingfacehub_api_token = API_TOKEN,
            repo_id=llm_model,
            temperature = temperature,
            max_new_tokens = max_tokens,
            top_k = top_k,
        )

    memory = ConversationBufferMemory(
        memory_key="chat_history",
        output_key='answer',
        return_messages=True
    )
        
    qa_chain = ConversationalRetrievalChain.from_llm(
        llm,
        retriever=CustomRetriever(vectorstore=vdb, thold=thold),
        chain_type="stuff",
        memory=memory,
        return_source_documents=True,
        verbose=False,
    )
    return qa_chain



# Initialize LLM
def initialize_LLM(llm_temperature, max_tokens, top_k, vector_db, thold, progress=gr.Progress()):
    llm_name = "meta-llama/Meta-Llama-3-8B-Instruct" #"mistralai/Mistral-7B-Instruct-v0.2"   
    qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, top_k, vector_db, thold)
    return qa_chain 



def format_chat_history(chat_history):
    formatted_chat_history = []
    for user_message, bot_message in chat_history:
        formatted_chat_history.append(f"User: {user_message}")
        formatted_chat_history.append(f"Assistant: {bot_message}")
    return formatted_chat_history



def postprocess(response):
    try:
        result=response["answer"]
        #Here should be a binary classification model.
        if not "I don't know" in result:            
            for doc in response['source_documents']:
                file_doc="\n\nFile: " + doc.metadata["source"].split('/')[-1]
                page="\nPage: " + str(doc.metadata["page"])
                content="\nFragment: " + doc.page_content.strip()
                result+=file_doc+page+content
        return result
    except:
        return "I don't know."