ChatErector.py

from llm.utils import initialize_LLM, format_chat_history, postprocess
from db.utils import initialize_database
import gradio as gr
import spaces

def initializer(list_file_obj, llm_temperature, max_tokens, top_k, thold, progress=gr.Progress()):
    vdb=initialize_database(list_file_obj)
    qa_chain=initialize_LLM(llm_temperature, max_tokens, top_k, vdb, thold)#, progress)
    return qa_chain, "Success."


#spaces.GPU
def conversation(qa_chain, message, history):
    formatted_chat_history = format_chat_history(history)#message, history)
    # Generate response using QA chain
    response = qa_chain.invoke({"question": message, "chat_history": formatted_chat_history})
    response_answer = postprocess(response)#response["answer"]
    #if response_answer.find("Helpful Answer:") != -1:
        #response_answer = response_answer.split("Helpful Answer:")[-1]
    #response_sources = response["source_documents"]
    #response_source1 = response_sources[0].page_content.strip()
    #response_source2 = response_sources[1].page_content.strip()
    #response_source3 = response_sources[2].page_content.strip()
    # Langchain sources are zero-based
    #response_source1_page = response_sources[0].metadata["page"] + 1
    #response_source2_page = response_sources[1].metadata["page"] + 1
    #response_source3_page = response_sources[2].metadata["page"] + 1
    # Append user message and response to chat history
    new_history = history + [(message, response_answer)]
    return qa_chain, gr.update(value=""), new_history #, response_source1, response_source1_page, response_source2, response_source2_page, response_source3, response_source3_page