test_load.py

import torch
import sys
from transformers import AutoModel, AutoConfig

# Repository ID on Hugging Face Hub
REPO_ID = "StarNetLaboratory/mosaic"

# Since custom modeling code (foundation_bert.py) is used, 
# trust_remote_code must be set to True.
TRUST_CODE = True

print(f"--- 1. Attempting to load configuration ---")
try:
    # Attempt to load config to verify config.json is present and readable
    config = AutoConfig.from_pretrained(REPO_ID, trust_remote_code=TRUST_CODE)
    print(f"✅ Config loaded successfully: {config.architectures}")
except Exception as e:
    print(f"❌ Config loading failed: {e}")
    sys.exit(1)

print(f"\n--- 2. Attempting to load model ---")
try:
    # Detect device (GPU if available, else CPU)
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    
    # This call triggers transformers to download and execute foundation_bert.py
    # and load weights from model.safetensors
    model = AutoModel.from_pretrained(
        REPO_ID, 
        config=config, 
        trust_remote_code=TRUST_CODE, 
        torch_dtype=torch.float32 # Match the dtype used during training/local testing
    ).to(device)
    
    model.eval()
    
    # Calculate and print total parameters to verify the architecture
    total_params = sum(p.numel() for p in model.parameters())
    print(f"✅ Model loaded successfully! Total parameters: {total_params:,}")

except Exception as e:
    print(f"❌ Model loading failed.")
    print(f"Check file integrity and remote code logic (foundation_bert.py).")
    print(f"Error details: {e}")
    sys.exit(1)