trainer_state.json

{
  "best_metric": 1.82,
  "best_model_checkpoint": "/mnt/models/runs/sft-qwen2.5-7b-k8s-001/checkpoint-000120",
  "epoch": 0.40,
  "global_step": 120,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "step": 10,
      "loss": 2.91,
      "learning_rate": 0.00006,
      "epoch": 0.03
    },
    {
      "step": 60,
      "loss": 2.14,
      "learning_rate": 0.00018,
      "epoch": 0.20
    },
    {
      "step": 120,
      "eval_loss": 1.82,
      "epoch": 0.40
    }
  ],
  "save_steps": 120,
  "logging_steps": 10,
  "train_runtime": 842.31,
  "train_samples_per_second": 21.7
}