app.py

import streamlit as st
import torch
from transformers import AutoTokenizer
from semviqa.SER.qatc_model import QATCForQuestionAnswering

# Load QATC Model
@st.cache_resource()
def load_qatc_model():
    tokenizer = AutoTokenizer.from_pretrained("xuandin/semviqa-qatc-vimrc-viwikifc")
    model = QATCForQuestionAnswering.from_pretrained("xuandin/semviqa-qatc-vimrc-viwikifc")
    return tokenizer, model

# Streamlit UI Configuration
st.set_page_config(page_title="SemViQA Demo", layout="wide")

# Improved UI Design
st.markdown("""
    <style>
        .big-title {
            font-size: 36px;
            font-weight: bold;
            color: #4A90E2;
            text-align: center;
        }
        .sub-title {
            font-size: 20px;
            color: #666;
            text-align: center;
        }
        .stButton>button {
            background-color: #4CAF50;
            color: white;
            font-size: 16px;
            width: 100%;
            border-radius: 8px;
            padding: 10px;
        }
        .stTextArea textarea {
            font-size: 16px;
        }
        .result-box {
            background-color: #f9f9f9;
            padding: 20px;
            border-radius: 10px;
            box-shadow: 0px 4px 8px rgba(0, 0, 0, 0.1);
        }
    </style>
""", unsafe_allow_html=True)

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 a claim and context to verify its accuracy</p>", unsafe_allow_html=True)

# Sidebar - Configuration Settings
st.sidebar.header("⚙️ Settings")
tfidf_threshold = st.sidebar.slider("🔧 TF-IDF Threshold", 0.0, 1.0, 0.5, 0.01)
length_ratio_threshold = st.sidebar.slider("📏 Length Ratio Threshold", 0.1, 1.0, 0.5, 0.01)
qatc_model = st.sidebar.selectbox("🤖 Select QATC Model", ["xuandin/semviqa-qatc-vimrc-viwikifc"])

# User Input Fields
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.")

if st.button("🔎 Verify"):
    tokenizer, model = load_qatc_model()
    inputs = tokenizer(claim, context, return_tensors="pt", truncation=True, max_length=512)
    
    with torch.no_grad():
        outputs = model(**inputs)
    
    start_idx = torch.argmax(outputs.start_logits)
    end_idx = torch.argmax(outputs.end_logits)
    tokens = inputs["input_ids"][0][start_idx : end_idx + 1]
    evidence_result = tokenizer.decode(tokens, skip_special_tokens=True)
    
    st.markdown("""
        <div class='result-box'>
            <h3>📌 Result</h3>
            <p><strong>🔍 Evidence:</strong> {}</p>
        </div>
    """.format(evidence_result), unsafe_allow_html=True)