{
  "base_checkpoint": "North-ML1/Aurora-Proelia",
  "dataset": "open-r1/OpenR1-Math-220k default",
  "objective": "response-masked SFT on verified reasoning traces with Proelia chat replay",
  "reasoning_rows": 500,
  "replay_rows": 40,
  "device": "mps",
  "learning_rate": 2e-07,
  "eval_loss_before": 2.6482388231489393,
  "eval_loss_after": 2.3090534660551283,
  "elapsed_seconds": 988.8
}