training_metadata.json

{
  "epoch": 1,
  "base_model": "Qwen/Qwen2.5-14B-Instruct",
  "metrics": {
    "sft/loss_mean": 0.7318060994148254,
    "sft/loss_final": 0.7072445154190063,
    "sft/loss_std": 0.3951688552930663,
    "sft/grad_norm_mean": 8.714482615058538,
    "sft/grad_norm_max": 1520.0,
    "sft/tokens_trained": 3828602,
    "sft/samples_trained": 79264,
    "sft/optimizer_steps": 2477,
    "sft/loss_per_token": 0.0004725922114807613,
    "global_step": 2477,
    "epoch": 1
  }
}