training_metrics.json

{
  "steps": 69,
  "samples": 548,
  "dropped_samples": 0,
  "max_sequence_tokens": 1956,
  "supervised_tokens": 78740,
  "loss": 0.4510538172454649,
  "trainable_parameters": 2678784,
  "optimizer": "PagedAdamW8bit",
  "gradient_checkpointing": true,
  "attention": "auto",
  "attention_kernels": {
    "efficient": 38360
  },
  "optimizer_state_mib": 5.812744140625,
  "peak_cuda_gib": 8.748569011688232,
  "peak_cuda_reserved_gib": 13.013671875
}