app.py

import gradio as gr
import os
from dotenv import load_dotenv
load_dotenv()




# Use followin json data to feed to Chroma
import json
with open("data/processed/final_data_for_vectorstore.json",'r') as file:
    data4chroma= json.load(file)

# Initiate vector store
from langchain_community.vectorstores import Chroma
from langchain_huggingface import HuggingFaceEmbeddings
embedding_function=HuggingFaceEmbeddings(model_name='all-MiniLM-L6-v2')
vectorstore=Chroma.from_texts(texts=data4chroma['chunks'],
                             embedding=embedding_function,
                             ids=data4chroma["chunk_ids"],
                             metadatas=data4chroma["chunk_metadatas"],
                             collection_name='qual_books',
                             )


from langchain_core.prompts import ChatPromptTemplate
template="""You are a helpful AI assistant. Please answer the query based on provided context.\
         *Do not make any assumptions if you don't know the answer. In that case just respond by saying\
          the answer of query cannot be found in the given context.
         *The English of the provided text is not well-structured. You should respond with the same content but in improved, clear, and correct English, without simply copying the original text.
         *Also provide the response in bullet points but in detail where necessary.
    Context: {context}
    Query: {question}
    
    Answer:
         """
prompt= ChatPromptTemplate.from_template(template)

from langchain_huggingface import HuggingFaceEndpoint
llm=HuggingFaceEndpoint(repo_id="meta-llama/Meta-Llama-3.1-70B-Instruct",
                               max_new_tokens=3000,
                              top_k=20,  
                                top_p=0.95,  
                                typical_p=0.95,  
                                temperature=0.001,  
                                repetition_penalty=1.03,
                                huggingfacehub_api_token=os.getenv("huggingfacehub_api_token")
                                 )
chain = prompt | llm


def respond(
            query: str,
            data_type: str = "Preprocessed doc",
            llm_chain = chain,
            vectorstore=vectorstore
            ):
    """
    Generate a response to a user query using document retrieval and language model 
    completion
    Parameters:
    chatbot (List): List representing the chatbot's conversation history.
    message (str): The user's query.
    data_type (str): Type of data used for document retrieval
    temperature (float);
    Returns:
    Tuple: A tuple containing an empty string, the updated chat history,
    and reference from retrieved documents
    """
    # Retrieve embedding function from code env resources
    
    if data_type=="Preprocessed doc":
        retriever=vectorstore.as_retriever(search_type="mmr",
                              search_kwargs={"k":10,"fetch_k":100})
        retrieved_docs=retriever.invoke(query)


        input_2_chain={"context": retrieved_docs, "question":query}
        
        response=llm_chain.invoke(input_2_chain)
    return  response
    

demo = gr.Interface(fn=respond, inputs="text", outputs="text")
demo.launch(share=True)