{
  "encoder": "answerdotai/ModernBERT-large",
  "head_layers": 2,
  "max_len": 512,
  "head_max_len": 192,
  "max_prefixes": 6,
  "act_costs": {
    "escalate": 0.5
  },
  "cost_wrong_act": 3.0,
  "amp_dtype": "bf16",
  "model_name": "rl-agent",
  "temperature": [
    2.3200724124908447,
    1.0,
    1.0
  ],
  "training": {
    "updates": 7313,
    "epochs_completed": 1,
    "hours": 1.96,
    "world_size": 1,
    "fine_tuned_from_checkpoint": true,
    "laya_train": {
      "epochs": 4,
      "micro_batch": 8,
      "grad_accum": 8,
      "encoder_lr": 2.5e-05,
      "head_lr": 0.0001,
      "min_lr": 1e-06,
      "weight_decay": 0.01,
      "grad_clip": 1.0,
      "loss": "rlcd",
      "rl_samples": 4,
      "sigma_start": 0.4,
      "sigma_end": 0.1,
      "w_sph": 0.75,
      "w_rps": 1.0,
      "shuffle_options": [],
      "option_layout": null,
      "freeze_encoder": false,
      "calib_max": 400,
      "calib_frac": 0.1,
      "calib_seed": 20260922,
      "seed": 0,
      "max_len": null,
      "head_max_len": null,
      "amp": null,
      "gradient_checkpointing": null,
      "log_every": 100,
      "label_smoothing": 0.0,
      "text_column": "text",
      "label_column": "label",
      "question_id": "label",
      "instructions": null,
      "eval_data": null,
      "target_error": 0.1,
      "min_abstain_n": 10
    },
    "laya_train_calibration": {
      "choice": {
        "items": 197,
        "temperature": 2.3200724124908447,
        "issues": [],
        "accuracy": 0.700507614213198,
        "mean_confidence": 0.7349079253696408
      },
      "score": {
        "items": 0,
        "temperature": 1.0,
        "issues": []
      },
      "noul": {
        "items": 203,
        "temperature": 1.0,
        "issues": [
          "the fit came back unchanged at 1.0 (accuracy 1.00, mean confidence 1.000 on the calibration items), which usually means they gave it nothing to correct; this type's confidences are uncalibrated"
        ],
        "accuracy": 1.0,
        "mean_confidence": 0.9999999999981326
      }
    },
    "train_report": {
      "eval_source": "calibration_slice",
      "eval_mode": "in_sample_calibration",
      "is_held_out": false,
      "eval_items": 400,
      "note": "Metrics reflect in-sample calibration fit, not independent held-out evaluation. Provide --eval for held-out validation.",
      "before": {
        "items": 400,
        "loss": 1.6005,
        "accuracy": 0.4,
        "mean_confidence": 0.6787,
        "ece": 0.2787,
        "brier": 0.7818,
        "brier_top1": 0.3007,
        "by_type": {
          "choice": {
            "items": 197,
            "loss": 2.5345,
            "accuracy": 0.2081,
            "mean_confidence": 0.6239,
            "ece": 0.4157,
            "brier": 1.079,
            "brier_top1": 0.3564
          },
          "noul": {
            "items": 203,
            "loss": 0.6941,
            "accuracy": 0.5862,
            "mean_confidence": 0.732,
            "ece": 0.1485,
            "brier": 0.4934,
            "brier_top1": 0.2467
          }
        }
      },
      "after": {
        "items": 400,
        "loss": 0.4219,
        "accuracy": 0.8525,
        "mean_confidence": 0.8694,
        "ece": 0.0372,
        "brier": 0.2124,
        "brier_top1": 0.0912,
        "by_type": {
          "choice": {
            "items": 197,
            "loss": 0.8566,
            "accuracy": 0.7005,
            "mean_confidence": 0.7349,
            "ece": 0.0755,
            "brier": 0.4313,
            "brier_top1": 0.1852
          },
          "noul": {
            "items": 203,
            "loss": 0.0,
            "accuracy": 1.0,
            "mean_confidence": 1.0,
            "ece": 0.0,
            "brier": 0.0,
            "brier_top1": 0.0
          }
        }
      },
      "comparison": {
        "delta_accuracy": 0.4525,
        "delta_loss": -1.1786,
        "delta_ece": -0.2415,
        "delta_brier": -0.5694,
        "delta_brier_top1": -0.2095,
        "delta_mean_confidence": 0.1907
      },
      "calibration": {
        "choice": {
          "items": 197,
          "temperature": 2.3200724124908447,
          "issues": [],
          "accuracy": 0.700507614213198,
          "mean_confidence": 0.7349079253696408
        },
        "score": {
          "items": 0,
          "temperature": 1.0,
          "issues": []
        },
        "noul": {
          "items": 203,
          "temperature": 1.0,
          "issues": [
            "the fit came back unchanged at 1.0 (accuracy 1.00, mean confidence 1.000 on the calibration items), which usually means they gave it nothing to correct; this type's confidences are uncalibrated"
          ],
          "accuracy": 1.0,
          "mean_confidence": 0.9999999999981326
        }
      },
      "abstention_thresholds": {
        "choice:6-10": 0.8826,
        "noul:2": 1.0
      },
      "training": {
        "epochs": 4,
        "micro_batch": 8,
        "grad_accum": 8,
        "encoder_lr": 2.5e-05,
        "head_lr": 0.0001,
        "min_lr": 1e-06,
        "weight_decay": 0.01,
        "grad_clip": 1.0,
        "loss": "rlcd",
        "rl_samples": 4,
        "sigma_start": 0.4,
        "sigma_end": 0.1,
        "w_sph": 0.75,
        "w_rps": 1.0,
        "shuffle_options": [],
        "option_layout": null,
        "freeze_encoder": false,
        "calib_max": 400,
        "calib_frac": 0.1,
        "calib_seed": 20260922,
        "seed": 0,
        "max_len": null,
        "head_max_len": null,
        "amp": null,
        "gradient_checkpointing": null,
        "log_every": 100,
        "label_smoothing": 0.0,
        "text_column": "text",
        "label_column": "label",
        "question_id": "label",
        "instructions": null,
        "eval_data": null,
        "target_error": 0.1,
        "min_abstain_n": 10
      }
    },
    "abstention_thresholds": {
      "choice:6-10": 0.8826,
      "noul:2": 1.0
    }
  },
  "fine_tuned": true
}