app.py

import streamlit as st
import pandas as pd
from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
from langchain_community.vectorstores import FAISS
from langchain_huggingface import HuggingFaceEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.schema import Document
from sentence_transformers import SentenceTransformer
from langchain_community.document_loaders import DirectoryLoader, TextLoader


model_name = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
embedding_model_name = "sentence-transformers/all-MiniLM-L6-v2"


docs = []
db = None
extracted_text = None


# ------------------------------
# Load Model for pretraining
# ------------------------------
@st.cache_resource
def load_model():
    # model_name = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForCausalLM.from_pretrained(model_name)
    return pipeline("text-generation", model=model, tokenizer=tokenizer)

@st.cache_resource
def extract_text():
    uploaded_data_path = "./ramayan"
    loader = DirectoryLoader(
        path=uploaded_data_path,
        glob="**/*.txt",  
        loader_cls=TextLoader,
        recursive=True 
    )
    documents = loader.load()
    return "\n".join([doc.page_content for doc in documents])

def build_faiss(_docs):
    embeddings = HuggingFaceEmbeddings(model_name=embedding_model_name)
    return FAISS.from_documents(_docs, embeddings)


with st.spinner("🔄 Loading Model..."):
    generator = load_model()
with st.spinner("🔄 Loading Knowldge Base..."):
    extracted_text = extract_text()
    splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
    docs = [Document(page_content=chunk) for chunk in splitter.split_text(extracted_text)]
    db = build_faiss(docs)
    st.success("✅ Knowledge Base ready! From :- Ramayan")


# ------------------------------
# Title
# ------------------------------
st.title("📚 RAG For Ramayan")
st.markdown("This app uses a local LLM model to answer questions about Ramayan using RAG (Retrieval Augmented Generation).")   

with st.form(key='my_form'):
    query = st.text_input("💬 Ask a question about Ramayan(Required)", placeholder="Rishyasring's Departure")
    max_new_tokens_model = st.slider("Max New Tokens (Optional):", min_value=50, max_value=500, value=150, step=25)
    temperature_model = st.slider("Temperature (Optional):", min_value=0.0, max_value=0.9, value=0.5, step=0.1)
    submit_button = st.form_submit_button("Submit")

if submit_button:
    if query and db and extracted_text and len(docs) > 0:
        retriever = db.as_retriever(search_kwargs={"k": 3})
        retrieved_docs = retriever.get_relevant_documents(query)
        context = "\n".join([doc.page_content for doc in retrieved_docs])

        with st.spinner("🤔 Generating answer..."):
            result = generator(
                f"Context:\n{context}\n\nQuestion: {query}\nAnswer:",
                max_new_tokens=max_new_tokens_model,
                temperature=temperature_model,
                top_p=0.9
            )

            generated = result[0]["generated_text"]
            answer_only = generated.split("Answer:")[-1].strip()

        st.write("📝 Answer:", answer_only)