model_service.py

import time
from pathlib import Path
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch
from logger import logger
from config import CACHE_DIR, MODEL_NAME


tokenizer = None
model = None
generator = None
model_loaded = False
startup_time = time.time()

def is_model_cached(model_name: str) -> bool:
    try:
        model_path = Path(CACHE_DIR) / f"models--{model_name.replace('/', '--')}"
        return model_path.exists() and any(model_path.iterdir())
    except Exception as e:
        logger.error(f"Error checking cache: {e}")
        return False

def load_model():
    global tokenizer, model, generator, model_loaded

    try:
        logger.info(f"Loading model: {MODEL_NAME}")
        logger.info(f"CUDA available: {torch.cuda.is_available()}")
        start = time.time()

        tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, cache_dir=CACHE_DIR)
        if tokenizer.pad_token is None:
            tokenizer.pad_token = tokenizer.eos_token

        model = AutoModelForCausalLM.from_pretrained(
            MODEL_NAME,
            cache_dir=CACHE_DIR,
            torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
            device_map="auto" if torch.cuda.is_available() else None,
            low_cpu_mem_usage=True,
        )

        device = 0 if torch.cuda.is_available() else -1
        generator = pipeline(
            "text-generation",
            model=model,
            tokenizer=tokenizer,
            device=device,
        )

        model_loaded = True
        logger.info(f"✅ Model loaded in {time.time()-start:.2f}s on {model.device}")
        return True

    except Exception as e:
        logger.error(f"❌ Error loading model: {e}", exc_info=True)
        model_loaded = False
        return False

def generate_response(message: str, max_length: int, temperature: float, top_p: float):
    if not generator:
        return "❌ Model not loaded yet", 0.0

    start = time.time()
    try:
        result = generator(
            message,
            max_length=max_length,
            temperature=temperature,
            top_p=top_p,
            num_return_sequences=1,
            pad_token_id=tokenizer.eos_token_id,
            do_sample=True,
            repetition_penalty=1.1
        )
        text = result[0]['generated_text']
        reply = text[len(message):].strip() if text.startswith(message) else text.strip()
        if not reply:
            reply = "I'm not sure how to respond to that."
        return reply, time.time()-start

    except Exception as e:
        logger.error(f"Generation error: {e}", exc_info=True)
        return f"❌ Error: {e}", 0.0