app.py

from transformers import M2M100ForConditionalGeneration, M2M100Tokenizer
import gradio as gr

# Load M2M100 multilingual model (supports 100+ languages)
model_name = "facebook/m2m100_418M"
tokenizer = M2M100Tokenizer.from_pretrained(model_name)
model = M2M100ForConditionalGeneration.from_pretrained(model_name)

def translate_to_english(text):
    # Detect language (roughly by script, can be improved later)
    if any("\u0600" <= c <= "\u06FF" for c in text):   # Arabic/Urdu script
        lang = "ar"  # works for Urdu too, since M2M100 uses ISO codes
    elif any("\u0900" <= c <= "\u097F" for c in text): # Hindi (Devanagari)
        lang = "hi"
    else:
        lang = "en"

    # If already English, return as is
    if lang == "en":
        return f"🗣 You: {text}\n🌐 Detected: English\n✅ Translation: {text}"

    # Set tokenizer to source lang
    tokenizer.src_lang = lang
    encoded = tokenizer(text, return_tensors="pt")
    generated_tokens = model.generate(**encoded, forced_bos_token_id=tokenizer.get_lang_id("en"))
    translation = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]

    return f"🗣 You: {text}\n🌐 Detected: {lang.upper()}\n✅ Translation: {translation}"

# ---- Gradio UI ----
with gr.Blocks(css=".gradio-container {font-family: 'Poppins', sans-serif;}") as demo:
    gr.Markdown("## 🌍 Multilingual → English Chatbot (Arabic, Urdu, Hindi, English)")

    with gr.Row():
        inp = gr.Textbox(placeholder="Type something here...", label="Your Message")
    out = gr.Textbox(label="Chat Response", interactive=False)

    inp.submit(translate_to_english, inp, out)

demo.launch(server_name="0.0.0.0", server_port=7860)