app.py

import os

os.environ["HF_HOME"] = "/app/hf_cache"  # Or any writable folder in your container
from transformers import AutoTokenizer

from flask import Flask, request
from transformers import AutoTokenizer, AutoModelForCausalLM
from gtts import gTTS
from twilio.twiml.messaging_response import MessagingResponse
import torch

app = Flask(__name__)

model_id = "sarvamai/sarvam-m"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_in_4bit=True)

@app.route("/")
def home():
    return "Sarvam WhatsApp Malayalam Chatbot is running!"

@app.route("/whatsapp", methods=["POST"])
def whatsapp():
    msg = request.form.get('Body')
    sender = request.form.get('From')

    inputs = tokenizer(msg, return_tensors="pt").to(model.device)
    outputs = model.generate(**inputs, max_new_tokens=100)
    reply = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Convert reply to Malayalam speech (optional audio hosting needed)
    tts = gTTS(reply, lang="ml")
    tts.save("reply.mp3")

    resp = MessagingResponse()
    resp.message(reply)
    return str(resp)

if __name__ == "__main__":
    app.run(host="0.0.0.0", port=7860)