training_metadata.json

{
  "epoch": 3,
  "base_model": "Qwen/Qwen2.5-14B-Instruct",
  "metrics": {
    "sft/loss_mean": 0.5125790238380432,
    "sft/loss_final": 0.5577956438064575,
    "sft/loss_std": 0.04738047259057726,
    "sft/grad_norm_mean": 1.484967955187727,
    "sft/grad_norm_max": 1.703125,
    "sft/tokens_trained": 3828496,
    "sft/samples_trained": 79264,
    "sft/optimizer_steps": 2477,
    "sft/loss_per_token": 0.00033085525851376965,
    "global_step": 7431,
    "epoch": 3
  }
}