app.py

import streamlit as st
import torch
import os
import time
from threading import Thread
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
from langchain_community.document_loaders import PyPDFLoader, TextLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import FAISS
from langchain.retrievers import BM25Retriever, EnsembleRetriever
from langchain.schema import Document
from langchain.docstore.document import Document as LangchainDocument

# --- HF Token ---
HF_TOKEN = st.secrets["HF_TOKEN"]

# --- Page Config ---
st.set_page_config(page_title="DigiTwin RAG", page_icon="📂", layout="centered")
st.title("📂 DigiTs the Twin")

# --- Sidebar ---
with st.sidebar:
    st.header("📄 Upload Knowledge Files")
    uploaded_files = st.file_uploader("Upload PDFs or .txt files", accept_multiple_files=True, type=["pdf", "txt"])
    hybrid_toggle = st.checkbox("🔀 Enable Hybrid Search", value=True)
    clear_chat = st.button("🧹 Clear Chat History")

# --- Session State ---
if "messages" not in st.session_state or clear_chat:
    st.session_state.messages = []

# --- Load Model + Tokenizer ---
@st.cache_resource
def load_model():
    model_id = "tiiuae/falcon-7b-instruct"
    tokenizer = AutoTokenizer.from_pretrained(model_id, token=HF_TOKEN)
    model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto", token=HF_TOKEN)
    return tokenizer, model

tokenizer, model = load_model()

# --- Process Documents ---
def process_documents(files):
    documents = []
    for file in files:
        if file.name.endswith(".pdf"):
            loader = PyPDFLoader(file)
        else:
            loader = TextLoader(file)
        docs = loader.load()
        documents.extend(docs)
    return documents

def chunk_documents(documents):
    splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
    return splitter.split_documents(documents)

# --- Build Hybrid Retriever ---
def build_retrievers(chunks):
    embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
    faiss_vectorstore = FAISS.from_documents(chunks, embeddings)
    faiss_retriever = faiss_vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 5})
    bm25_retriever = BM25Retriever.from_documents([LangchainDocument(page_content=d.page_content) for d in chunks])
    bm25_retriever.k = 5
    hybrid = EnsembleRetriever(retrievers=[faiss_retriever, bm25_retriever], weights=[0.5, 0.5])
    return faiss_retriever, hybrid

# --- Inference with Streaming ---
def generate_stream_response(system_prompt):
    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
    inputs = tokenizer(system_prompt, return_tensors="pt").to(model.device)
    generation_kwargs = dict(**inputs, streamer=streamer, max_new_tokens=300)
    thread = Thread(target=model.generate, kwargs=generation_kwargs)
    thread.start()
    partial_output = ""
    for token in streamer:
        partial_output += token
        yield partial_output

# --- Main App Logic ---
if uploaded_files:
    with st.spinner("Processing documents..."):
        docs = process_documents(uploaded_files)
        chunks = chunk_documents(docs)
        faiss_retriever, hybrid_retriever = build_retrievers(chunks)
        retriever = hybrid_retriever if hybrid_toggle else faiss_retriever
        st.success("Knowledge base ready. Ask your question below.")

    for msg in st.session_state.messages:
        with st.chat_message(msg["role"]):
            st.markdown(msg["content"])

    user_input = st.chat_input("💬 Ask DigiTwin something...")
    if user_input:
        st.chat_message("user").markdown(user_input)
        st.session_state.messages.append({"role": "user", "content": user_input})

        with st.chat_message("assistant"):
            context_docs = retriever.get_relevant_documents(user_input)
            context_text = "\n".join([doc.page_content for doc in context_docs])

            system_prompt = (
                "You are DigiTwin, an expert advisor in asset integrity, reliability, inspection, and maintenance "
                "of topside piping, structural, mechanical systems, floating units, pressure vessels (VII), and pressure safety devices (PSD's).\n\n"
                f"Context:\n{context_text}\n\n"
                f"User: {user_input}\nAssistant:"
            )

            full_response = ""
            response_area = st.empty()
            for partial_output in generate_stream_response(system_prompt):
                full_response = partial_output
                response_area.markdown(full_response)
            st.session_state.messages.append({"role": "assistant", "content": full_response})
else:
    st.info("👈 Upload one or more PDFs or .txt files to begin.")