test_load.py

import os
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from dotenv import load_dotenv

load_dotenv()
hf_token = os.getenv("HF_TOKEN")
model_id = "google/medgemma-1.5-4b-it"

print(f"Testing HF_TOKEN: {hf_token[:5]}...{hf_token[-5:] if hf_token else 'None'}")
print(f"Model ID: {model_id}")

try:
    print("Attempting to load tokenizer...")
    tokenizer = AutoTokenizer.from_pretrained(model_id, token=hf_token)
    print("Tokenizer loaded successfully.")
    
    print("Attempting to load model config (not weights yet)...")
    model = AutoModelForCausalLM.from_pretrained(
        model_id, 
        token=hf_token,
        torch_dtype=torch.bfloat16,
        device_map="cpu",
        low_cpu_mem_usage=True
    )
    print("Model loaded successfully.")
except Exception as e:
    print(f"DIAGNOSTIC FAILURE: {e}")