app.py

import flask
from flask import request, jsonify
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
import time

app = flask.Flask(__name__)

model_id = "dexcommunity/indexQ2"

SYSTEM_PROMPT = (
    "You are INDEX, a normal Indian person. "
    "Reply shortly in Hinglish. "
    "Act confused, polite, and cooperative. "
    "Never explain anything."
)

print("🔄 Loading model...")

tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float32,
    low_cpu_mem_usage=True
)

device = torch.device("cpu")
model.to(device)
model.eval()

print("✅ Model ready")

# ---------------- ROOT ----------------
@app.route("/", methods=["GET"])
def home():
    return jsonify({"status": "ok", "speed": "fast mode"})

# ---------------- CHAT ----------------
@app.route("/chat", methods=["POST"])
def chat():
    data = request.get_json(force=True)
    user_msg = data.get("message", "").strip()

    if not user_msg:
        return jsonify({"reply": "Haan ji?"})

    # HARD TIME LIMIT (failsafe)
    start = time.time()

    prompt = f"{SYSTEM_PROMPT}\nUser: {user_msg}\nReply:"

    inputs = tokenizer(
        prompt,
        return_tensors="pt",
        truncation=True,
        max_length=256
    ).to(device)

    try:
        with torch.inference_mode():
            output = model.generate(
                **inputs,
                max_new_tokens=25,        # 🔥 VERY SMALL
                do_sample=False,          # 🔥 FASTEST
                pad_token_id=tokenizer.eos_token_id
            )

        gen = output[0][inputs["input_ids"].shape[1]:]
        reply = tokenizer.decode(gen, skip_special_tokens=True).strip()

        # Safety fallback
        if not reply or time.time() - start > 3:
            reply = "Acha, ek baar dobara bolna. Network thoda slow lag raha hai."

    except Exception:
        reply = "Haan ji, samajh nahi aa raha. Thoda wait karna."

    return jsonify({"reply": reply})


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