app.py

import streamlit as st
import torch
from transformers import AutoTokenizer
from semviqa.ser.qatc_model import QATCForQuestionAnswering
from semviqa.tvc.model import ClaimModelForClassification
from semviqa.ser.ser_eval import extract_evidence_tfidf_qatc
from semviqa.tvc.tvc_eval import classify_claim
import time
import pandas as pd

# Load models with caching
@st.cache_resource()
def load_model(model_name, model_class, is_bc=False):
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = model_class.from_pretrained(model_name, num_labels=3 if not is_bc else 2)
    model.eval()
    return tokenizer, model

# Set up page configuration
st.set_page_config(page_title="SemViQA Demo", layout="wide")

# Custom CSS and JavaScript to make the sidebar sticky
st.markdown(
    """
    <style>
    /* Fix the sidebar */
    .stSidebar {
        position: fixed;
        top: 0;
        height: 100vh;
        overflow-y: auto;
        z-index: 1000;
    }
    /* Adjust main content to avoid overlap with the fixed sidebar */
    .main .block-container {
        margin-left: 25rem; /* Adjust this value based on your sidebar width */
    }
    </style>
    <script>
    // JavaScript to make the sidebar sticky
    window.addEventListener('scroll', function() {
        const sidebar = document.querySelector('.stSidebar');
        if (sidebar) {
            sidebar.style.top = `${window.scrollY}px`;
        }
    });
    </script>
    """,
    unsafe_allow_html=True,
)

# Container for the whole content with dynamic height
with st.container():
    st.markdown("<p class='big-title'>SemViQA: A Semantic Question Answering System for Vietnamese Information Fact-Checking</p>", unsafe_allow_html=True)
    st.markdown("<p class='sub-title'>Enter the claim and context to verify its accuracy</p>", unsafe_allow_html=True)

    # Sidebar: Global Settings
    with st.sidebar.expander("⚙️ Settings", expanded=True):
        tfidf_threshold = st.slider("TF-IDF Threshold", 0.0, 1.0, 0.5, 0.01)
        length_ratio_threshold = st.slider("Length Ratio Threshold", 0.1, 1.0, 0.5, 0.01)
        qatc_model_name = st.selectbox("QATC Model", [
            "SemViQA/qatc-infoxlm-viwikifc",
            "SemViQA/qatc-infoxlm-isedsc01",
            "SemViQA/qatc-vimrc-viwikifc",
            "SemViQA/qatc-vimrc-isedsc01"
        ])
        bc_model_name = st.selectbox("Binary Classification Model", [
            "SemViQA/bc-xlmr-viwikifc",
            "SemViQA/bc-xlmr-isedsc01",
            "SemViQA/bc-infoxlm-viwikifc",
            "SemViQA/bc-infoxlm-isedsc01",
            "SemViQA/bc-erniem-viwikifc",
            "SemViQA/bc-erniem-isedsc01"
        ])
        tc_model_name = st.selectbox("3-Class Classification Model", [
            "SemViQA/tc-xlmr-viwikifc",
            "SemViQA/tc-xlmr-isedsc01",
            "SemViQA/tc-infoxlm-viwikifc",
            "SemViQA/tc-infoxlm-isedsc01",
            "SemViQA/tc-erniem-viwikifc",
            "SemViQA/tc-erniem-isedsc01"
        ])
        show_details = st.checkbox("Show Probability Details", value=False)

    # Store verification history
    if 'history' not in st.session_state:
        st.session_state.history = []
    if 'latest_result' not in st.session_state:
        st.session_state.latest_result = None

    # Load the selected models
    tokenizer_qatc, model_qatc = load_model(qatc_model_name, QATCForQuestionAnswering)
    tokenizer_bc, model_bc = load_model(bc_model_name, ClaimModelForClassification, is_bc=True)
    tokenizer_tc, model_tc = load_model(tc_model_name, ClaimModelForClassification)

    # Icons for results
    verdict_icons = {
        "SUPPORTED": "✅",
        "REFUTED": "❌",
        "NEI": "⚠️"
    }

    # Tabs: Verify, History, About
    tabs = st.tabs(["Verify", "History", "About"])

    # --- Tab Verify ---
    with tabs[0]:
        st.subheader("Verify a Claim")
        # 2-column layout: input on the left, results on the right
        col_input, col_result = st.columns([2, 1])

        with col_input:
            claim = st.text_area("Enter Claim", "Vietnam is a country in Southeast Asia.")
            context = st.text_area("Enter Context", "Vietnam is a country located in Southeast Asia, covering an area of over 331,000 km² with a population of more than 98 million people.")
            verify_button = st.button("Verify", key="verify_button")

        with col_result:
            st.markdown("<h3>Verification Result</h3>", unsafe_allow_html=True)
            if verify_button:
                # Placeholder for displaying result/loading
                with st.spinner("Verifying..."):  # Thêm spinner khi đang xử lý
                    start_time = time.time()  # Bắt đầu đo thời gian inference

                    with torch.no_grad():
                        # Extract evidence
                        evidence_start_time = time.time()
                        evidence = extract_evidence_tfidf_qatc(
                            claim, context, model_qatc, tokenizer_qatc,
                            "cuda" if torch.cuda.is_available() else "cpu",
                            confidence_threshold=tfidf_threshold,
                            length_ratio_threshold=length_ratio_threshold
                        )
                        evidence_time = time.time() - evidence_start_time

                        # Classify the claim
                        verdict_start_time = time.time()
                        verdict = "NEI"
                        details = ""
                        prob3class, pred_tc = classify_claim(
                            claim, evidence, model_tc, tokenizer_tc,
                            "cuda" if torch.cuda.is_available() else "cpu"
                        )
                        if pred_tc != 0:
                            prob2class, pred_bc = classify_claim(
                                claim, evidence, model_bc, tokenizer_bc,
                                "cuda" if torch.cuda.is_available() else "cpu"
                            )
                            if pred_bc == 0:
                                verdict = "SUPPORTED"
                            elif prob2class > prob3class:
                                verdict = "REFUTED"
                            else:
                                verdict = ["NEI", "SUPPORTED", "REFUTED"][pred_tc]
                            if show_details:
                                details = f"""
                                    <p><strong>3-Class Probability:</strong> {prob3class.item():.2f}</p>
                                    <p><strong>3-Class Predicted Label:</strong> {['NEI', 'SUPPORTED', 'REFUTED'][pred_tc]}</p>
                                    <p><strong>2-Class Probability:</strong> {prob2class.item():.2f}</p>
                                    <p><strong>2-Class Predicted Label:</strong> {['SUPPORTED', 'REFUTED'][pred_bc]}</p>
                                """
                        verdict_time = time.time() - verdict_start_time

                        # Store verification history and the latest result
                        st.session_state.history.append({
                            "claim": claim,
                            "evidence": evidence,
                            "verdict": verdict,
                            "evidence_time": evidence_time,
                            "verdict_time": verdict_time,
                            "details": details
                        })
                        st.session_state.latest_result = {
                            "claim": claim,
                            "evidence": evidence,
                            "verdict": verdict,
                            "evidence_time": evidence_time,
                            "verdict_time": verdict_time,
                            "details": details
                        }

                        if torch.cuda.is_available():
                            torch.cuda.empty_cache()

                    # Display the result after verification
                    res = st.session_state.latest_result
                    st.markdown(f"""
                        <div class='result-box'>
                            <p><strong>Claim:</strong> {res['claim']}</p>
                            <p><strong>Evidence:</strong> {res['evidence']}</p>
                            <p><strong>Evidence Inference Time:</strong> {res['evidence_time']:.2f} seconds</p>
                            <p><strong>Verdict Inference Time:</strong> {res['verdict_time']:.2f} seconds</p>
                            <p class='verdict'><span class='verdict-icon'>{verdict_icons.get(res['verdict'], '')}</span>{res['verdict']}</p>
                            {res['details']}
                        </div>
                    """, unsafe_allow_html=True)
                    # Download Verification Result Feature
                    result_text = f"Claim: {res['claim']}\nEvidence: {res['evidence']}\nVerdict: {res['verdict']}\nDetails: {res['details']}"
                    st.download_button("Download Result", data=result_text, file_name="verification_result.txt", mime="text/plain")
            else:
                st.info("No verification result yet.")

    # --- Tab History ---
    with tabs[1]:
        st.subheader("Verification History")
        if st.session_state.history:
            # Convert history to DataFrame for easy download
            history_df = pd.DataFrame(st.session_state.history)
            st.download_button(
                label="Download Full History",
                data=history_df.to_csv(index=False).encode('utf-8'),
                file_name="verification_history.csv",
                mime="text/csv",
            )
            for idx, record in enumerate(reversed(st.session_state.history), 1):
                st.markdown(f"**{idx}. Claim:** {record['claim']}  \n**Result:** {verdict_icons.get(record['verdict'], '')} {record['verdict']}")
        else:
            st.write("No verification history yet.")

    # --- Tab About ---
    with tabs[2]:
        st.subheader("About")
        st.markdown("""
            <p align="center">
                <a href="https://arxiv.org/abs/2503.00955">
                    <img src="https://img.shields.io/badge/arXiv-2411.00918-red?style=flat&label=arXiv">
                </a>
                <a href="https://huggingface.co/SemViQA">
                    <img src="https://img.shields.io/badge/Hugging%20Face-Model-yellow?style=flat">
                </a>
                <a href="https://pypi.org/project/SemViQA">
                    <img src="https://img.shields.io/pypi/v/SemViQA?color=blue&label=PyPI">
                </a>
                <a href="https://github.com/DAVID-NGUYEN-S16/SemViQA">
                    <img src="https://img.shields.io/github/stars/DAVID-NGUYEN-S16/SemViQA?style=social">
                </a>
            </p>
        """, unsafe_allow_html=True)
        st.markdown("""
            **Description:**  
            SemViQA is a Semantic QA system designed for fact verification in Vietnamese.  
            The system extracts evidence from the provided context and classifies claims as **SUPPORTED**, **REFUTED**, or **NEI** (Not Enough Information) using advanced models.
        """)