src/rag.py

from langchain_huggingface import HuggingFacePipeline
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
import torch

def get_llm():
    model_id = "Qwen/Qwen3-0.6B"
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    model = AutoModelForCausalLM.from_pretrained(
        model_id,
        device_map="auto",
        torch_dtype=torch.float16
    )
    
    pipe = pipeline(
        "text-generation",
        model=model,
        tokenizer=tokenizer,
        max_new_tokens=512,
        temperature=0.1
    )
    return HuggingFacePipeline(pipeline=pipe)

def rewrite_query(question, chat_history):
    history_text = ""
    for turn in chat_history[-3:]:  
        history_text += f"User: {turn['question']}\n"
        history_text += f"Assistant: {turn['answer'][:150]}\n"

    rewrite_prompt = """Given the conversation history and a follow-up 
question, rewrite the question into a standalone search query that will 
retrieve relevant information from a document.

Conversation History:
{history}

Follow-up Question: {question}

Standalone Search Query:"""

    prompt = ChatPromptTemplate.from_template(rewrite_prompt)
    llm = get_llm()
    chain = prompt | llm | StrOutputParser()

    rewritten = chain.invoke({
        "history": history_text if history_text else "(none)",
        "question": question,
    })
    return rewritten.strip()

def answer_with_memory(vectorstore, question, chat_history, k=6):
    search_query = rewrite_query(question, chat_history)
    retrieved_docs = vectorstore.similarity_search(search_query, k=k)
    
    context_parts = []
    for i, doc in enumerate(retrieved_docs, 1):
        context_parts.append(f"[Source {i}]\n{doc.page_content}")
    context = "\n\n".join(context_parts)

    history_text = ""
    for turn in chat_history:
        history_text += f"User: {turn['question']}\n"
        history_text += f"Assistant: {turn['answer']}\n"

    template = """You are a highly efficient and accurate AI assistant for a document Q&A system. 
You MUST answer the user's question directly based ONLY on the provided Context.

Conversation History:
{history}

Context:
{context}

Question: {question}

Answer:"""

    prompt = ChatPromptTemplate.from_template(template)
    llm = get_llm()
    chain = prompt | llm | StrOutputParser()

    answer = chain.invoke({
        "history": history_text,
        "context": context,
        "question": question, 
    })

    return answer, retrieved_docs