model_loader.py

import os
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

HF_KEY = os.getenv("HF_Key")
MODEL_NAME = "meta-llama/Llama-3.2-1B-Instruct"

def load_model():
    print("Đang load model:", MODEL_NAME)

    tokenizer = AutoTokenizer.from_pretrained(
        MODEL_NAME,
        token=HF_KEY
    )

    model = AutoModelForCausalLM.from_pretrained(
        MODEL_NAME,
        torch_dtype=torch.float32,   # CPU nên để float32
        low_cpu_mem_usage=True,
        device_map="cpu",
        token=HF_KEY
    )

    print("Model loaded thành công")
    return model, tokenizer