scaling_1784740624/scaling_summary.json

{
  "seed": 20260722,
  "seq_len": 512,
  "eval_samples": 4,
  "epochs_cap": 1,
  "rows": [
    {
      "arm": "scale_80",
      "samples": 80,
      "points": 40960,
      "best_val_loss": 1.6425557136535645,
      "Affiliation-F": 0.06284466860196325,
      "F1_T": 0.06284466860196325,
      "Standard-F1": 0.06284466860196325,
      "VUS-PR": 0.024748325378238005
    },
    {
      "arm": "scale_120",
      "samples": 120,
      "points": 61440,
      "best_val_loss": 1.518439769744873,
      "Affiliation-F": 0.03773111238831465,
      "F1_T": 0.03773111238831465,
      "Standard-F1": 0.03773111238831465,
      "VUS-PR": 0.0173489950652482
    },
    {
      "arm": "scale_180",
      "samples": 180,
      "points": 92160,
      "best_val_loss": 1.6177384853363037,
      "Affiliation-F": 0.036136408553992104,
      "F1_T": 0.036136408553992104,
      "Standard-F1": 0.036136408553992104,
      "VUS-PR": 0.01593423381586978
    }
  ],
  "elapsed_seconds": 16.0,
  "note": "Reduced scaling run: nested synthetic-RCD sample counts, checkpoint-free, same architecture/loss."
}