train_summary_report.json

{
  "final_checkpoint": "/home/seonho/arnold/dit_only_phase1_ckpt/checkpoints/100000",
  "last_grad_norm": 0.4453125,
  "last_loss_finite": true,
  "last_lr": 0.0,
  "last_policy_update_called": false,
  "last_scalars": {
    "loss": 0.01060287281870842
  },
  "loaded_pretrained_model_path": null,
  "method": "base",
  "phase": "phase2",
  "policy_type": "groot",
  "script_name": "train_groot_arnold_yaml.py",
  "steps_completed": 100000,
  "steps_requested": 100000
}