app.py

import streamlit as st
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# 1. Model identifier
MODEL_ID = "songhieng/khmer-mt5-summarization"

# 2. Load tokenizer (you can choose fast or slow; fast is the default)
@st.cache_resource
def load_tokenizer_and_model(model_id):
    tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=True)
    model = AutoModelForSeq2SeqLM.from_pretrained(model_id)
    return tokenizer, model

tokenizer, model = load_tokenizer_and_model(MODEL_ID)

# 3. Streamlit page config
st.set_page_config(
    page_title="Khmer Text Summarization",
    layout="wide",
    initial_sidebar_state="expanded"
)

# 4. App header
st.title("📝 Khmer Text Summarization")
st.write("Paste your Khmer text below and click **Summarize** to get a concise summary.")

# 5. Sidebar summarization settings
st.sidebar.header("Summarization Settings")
max_length = st.sidebar.slider(
    "Maximum summary length", 50, 300, 150, step=10
)
min_length = st.sidebar.slider(
    "Minimum summary length", 10, 100, 30, step=5
)
num_beams = st.sidebar.slider(
    "Beam search width", 1, 10, 4, step=1
)

# 6. Text input
user_input = st.text_area(
    "Enter Khmer text here…", 
    height=300, 
    placeholder="សូមវាយអត្ថបទខ្មែរនៅទីនេះ…"
)

# 7. Summarize button
if st.button("Summarize"):
    if not user_input.strip():
        st.warning("⚠️ Please enter some text to summarize.")
    else:
        with st.spinner("Generating summary…"):
            # Tokenize
            inputs = tokenizer(
                user_input,
                return_tensors="pt",
                truncation=True,
                padding="longest"
            )
            # Generate
            summary_ids = model.generate(
                **inputs,
                max_length=max_length,
                min_length=min_length,
                num_beams=num_beams,
                length_penalty=2.0,
                early_stopping=True
            )
            # Decode
            summary = tokenizer.decode(
                summary_ids[0], 
                skip_special_tokens=True
            )
        # Display
        st.subheader("🔖 Summary:")
        st.write(summary)