src/streamlit_app.py

import streamlit as st
import tempfile
import os
import numpy as np

# LangChain community loaders & FAISS
from langchain_community.document_loaders import (
    PyPDFLoader,
    TextLoader,
    UnstructuredWordDocumentLoader,
    CSVLoader
)
from langchain_community.vectorstores import FAISS
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_huggingface import HuggingFaceEndpoint
from langchain.embeddings import HuggingFaceEmbeddings
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough

st.set_page_config(page_title="Ask RAG - HF Space", layout="wide")
st.title("Ask RAG - HuggingFace Space")

# HuggingFace API key (set via Space Secrets)
HF_TOKEN = os.environ.get("HUGGINGFACE_API_KEY")
if not HF_TOKEN:
    st.error("Please set the HuggingFace API key in your Space secrets!")
    st.stop()

# Wrapper embeddings via HF API
embeddings = HuggingFaceEmbeddings(
    model_name="intfloat/multilingual-e5-large-instruct",
    task="feature-extraction",
    model_kwargs={"use_auth_token": HF_TOKEN}
)

# Upload files
uploaded_files = st.file_uploader(
    "Upload files (PDF, DOCX, TXT, CSV)",
    type=["pdf", "docx", "txt", "csv"],
    accept_multiple_files=True
)

@st.cache_resource
def load_files(files):
    if not files:
        return None, []

    loaders = []
    temp_files = []

    for file in files:
        # Save temp file
        with tempfile.NamedTemporaryFile(delete=False, suffix=os.path.splitext(file.name)[-1]) as tmp:
            tmp.write(file.read())
            temp_files.append(tmp.name)

        # Choose loader
        if file.name.endswith(".pdf"):
            loaders.append(PyPDFLoader(tmp.name))
        elif file.name.endswith(".txt"):
            loaders.append(TextLoader(tmp.name))
        elif file.name.endswith(".docx"):
            loaders.append(UnstructuredWordDocumentLoader(tmp.name))
        elif file.name.endswith(".csv"):
            loaders.append(CSVLoader(tmp.name))

    # Load documents
    docs = []
    for loader in loaders:
        docs.extend(loader.load())

    # Split documents
    splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
    split_docs = splitter.split_documents(docs)

    # Create FAISS vectorstore
    vectorstore = FAISS.from_documents(split_docs, embeddings)

    return vectorstore, temp_files

vectorstore, temp_files = load_files(uploaded_files) if uploaded_files else (None, [])

if vectorstore:
    retriever = vectorstore.as_retriever()

    # Chat prompt
    chat_prompt = ChatPromptTemplate.from_template(
        """Use the context below to answer the question.

Context:
{context}

Question:
{question}"""
    )

    # LLM via HF Inference
    llm = HuggingFaceEndpoint(
        repo_id="AI-Sweden-Models/Llama-3-8B-instruct",
        task="text-generation",
        temperature=0.2,
        max_new_tokens=512,
        model_kwargs={"use_auth_token": HF_TOKEN}
    )

    # Build RAG chain
    rag_chain = (
        {
            "context": retriever | (lambda docs: "\n\n".join(d.page_content for d in docs)),
            "question": RunnablePassthrough(),
        }
        | chat_prompt
        | llm
    )

    # Session state
    if "messages" not in st.session_state:
        st.session_state.messages = []

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

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

        response = rag_chain.invoke(user_input)
        answer = response.content
        st.chat_message("assistant").markdown(answer)
        st.session_state.messages.append({"role": "assistant", "content": answer})

else:
    st.warning("Upload files to start querying.")

# Clear chat button
if st.button("Clear Chat"):
    st.session_state.messages = []
    for path in temp_files:
        try:
            os.remove(path)
        except:
            pass
    st.experimental_rerun()