abdurrouf14/QuestionAnswering

1import streamlit as st2import faiss3from io import BytesIO4from docx import Document5import numpy as np6from pypdf import PdfReader7 8from langchain_text_splitters import CharacterTextSplitter9from langchain_community.embeddings import HuggingFaceEmbeddings10from langchain_community.vectorstores import FAISS11from langchain_community.docstore.in_memory import InMemoryDocstore12from langchain_community.document_loaders import WebBaseLoader13 14from transformers import AutoTokenizer, AutoModelForQuestionAnswering, pipeline15 16# =========================17# Load QA model (CPU, no API key)18# =========================19@st.cache_resource20def load_qa_pipeline():21    model_name = "distilbert-base-uncased-distilled-squad"22    tokenizer = AutoTokenizer.from_pretrained(model_name)23    model = AutoModelForQuestionAnswering.from_pretrained(model_name)24    return pipeline(25        "question-answering",26        model=model,27        tokenizer=tokenizer,28        device=-1  # CPU29    )30 31qa_pipeline = load_qa_pipeline()32 33# =========================34# Build Vector Store35# =========================36def process_input(input_type, inputs):37    all_texts = []38 39    for data in inputs:40        # -------- Link --------41        if input_type == "Link":42            loader = WebBaseLoader(data)43            docs = loader.load()44            splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=100)45            chunks = splitter.split_documents(docs)46            texts = [c.page_content for c in chunks]47 48        # -------- PDF --------49        elif input_type == "PDF":50            reader = PdfReader(BytesIO(data.read()))51            text = "\n".join([p.extract_text() or "" for p in reader.pages])52            splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=100)53            texts = splitter.split_text(text)54 55        # -------- DOCX --------56        elif input_type == "DOCX":57            doc = Document(BytesIO(data.read()))58            text = "\n".join([p.text for p in doc.paragraphs])59            splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=100)60            texts = splitter.split_text(text)61 62        # -------- TXT / Text --------63        else:64            text = data if isinstance(data, str) else data.read().decode("utf-8")65            splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=100)66            texts = splitter.split_text(text)67 68        all_texts.extend(texts)69 70    # -------- Embeddings --------71    embeddings = HuggingFaceEmbeddings(72        model_name="sentence-transformers/all-mpnet-base-v2",73        model_kwargs={"device": "cpu"}74    )75 76    dim = len(embeddings.embed_query("test"))77    index = faiss.IndexFlatL2(dim)78 79    vectorstore = FAISS(80        embedding_function=embeddings.embed_query,81        index=index,82        docstore=InMemoryDocstore(),83        index_to_docstore_id={}84    )85    vectorstore.add_texts(all_texts)86 87    return vectorstore88 89# =========================90# Answer Question (FIXED)91# =========================92def answer_question(vectorstore, query):93    # ✅ FIX: use similarity_search instead of retriever.get_relevant_documents94    docs = vectorstore.similarity_search(query, k=4)95    context = " ".join([d.page_content for d in docs])96 97    if not context.strip():98        return "No relevant context found."99 100    result = qa_pipeline(101        question=query,102        context=context103    )104    return result["answer"]105 106# =========================107# Streamlit UI108# =========================109def main():110    st.set_page_config(page_title="RAG Q&A", layout="wide")111 112    st.markdown("## 🔍 RAG Question Answering System")113 114    input_type = st.selectbox(115        "Select Input Type",116        ["Direct Text", "Link", "PDF", "DOCX", "TXT"]117    )118 119    inputs = []120 121    # -------- Direct Text --------122    if input_type == "Direct Text":123        text = st.text_area("Paste your text here")124        if text.strip():125            inputs.append(text)126 127    # -------- Links --------128    elif input_type == "Link":129        n = st.number_input("Number of links", 1, 10, 1)130        for i in range(n):131            url = st.text_input(f"URL {i+1}")132            if url:133                inputs.append(url)134 135    # -------- Files --------136    else:137        types = {138            "PDF": ["pdf"],139            "DOCX": ["docx", "doc"],140            "TXT": ["txt"]141        }142        uploaded = st.file_uploader(143            "Upload files",144            type=types[input_type],145            accept_multiple_files=True146        )147        if uploaded:148            inputs.extend(uploaded)149 150    # -------- Build Vectorstore --------151    if st.button("Create Knowledge Base") and inputs:152        with st.spinner("Processing documents..."):153            st.session_state.vectorstore = process_input(input_type, inputs)154            st.success("Knowledge base ready!")155 156    # -------- Ask Question --------157    if "vectorstore" in st.session_state:158        query = st.text_input("Ask a question")159        if st.button("Get Answer") and query:160            with st.spinner("Thinking..."):161                answer = answer_question(st.session_state.vectorstore, query)162                st.markdown("### ✅ Answer")163                st.write(answer)164 165if __name__ == "__main__":166    main()167