train_results.json

{
    "epoch": 3.0,
    "max_memory_allocated (GB)": 63.2,
    "memory_allocated (GB)": 41.72,
    "total_memory_available (GB)": 94.62,
    "train_loss": NaN,
    "train_runtime": 480.8162,
    "train_samples_per_second": 35.55,
    "train_steps_per_second": 2.224
}