train_results.json

{
    "before_init_mem_cpu": 933711872,
    "before_init_mem_gpu": 0,
    "epoch": 10.0,
    "init_mem_cpu_alloc_delta": 2791841792,
    "init_mem_cpu_peaked_delta": 413900800,
    "init_mem_gpu_alloc_delta": 497994240,
    "init_mem_gpu_peaked_delta": 512,
    "train_loss": 1.525319737733178,
    "train_mem_cpu_alloc_delta": 8646197248,
    "train_mem_cpu_peaked_delta": 402849792,
    "train_mem_gpu_alloc_delta": 2010885120,
    "train_mem_gpu_peaked_delta": 15468135424,
    "train_runtime": 218631.5085,
    "train_samples": 3825056,
    "train_samples_per_second": 174.954,
    "train_steps_per_second": 0.021
}