model_loader.py

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from config import HF_TOKEN, MODEL_ID

def load_model():
    try:
        print(f"🔄 Loading tokenizer and model: {MODEL_ID}")
        
        tokenizer = AutoTokenizer.from_pretrained(
            MODEL_ID,
            token=HF_TOKEN or None,
            trust_remote_code=True
        )

        model = AutoModelForCausalLM.from_pretrained(
            MODEL_ID,
            token=HF_TOKEN or None,
            trust_remote_code=True,
            device_map="auto" if torch.cuda.is_available() else "cpu",
            torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
            low_cpu_mem_usage=True
        )

        print("✅ Model loaded successfully.")

        return pipeline(
            "text-generation",
            model=model,
            tokenizer=tokenizer,
            max_new_tokens=2048,
            do_sample=True,
            temperature=0.7,
            top_p=0.9
        )

    except Exception as e:
        print(f"❌ Failed to load model: {e}")
        raise RuntimeError(f"Model loading failed: {e}")