{ "base_beta": 0.5402383208274841, "base_bias": -0.4844930171966553, "kept_epoch": 3, "train": { "uncalibrated": { "nll": 0.13590236008167267, "brier": 0.03834998607635498, "ece": 0.008724160230938638, "accuracy": 0.9460732936859131 }, "global_scaling": { "nll": 0.1319091022014618, "brier": 0.036343060433864594, "ece": 0.024790527669744766, "accuracy": 0.9506661891937256 }, "calibrated": { "nll": 0.16371336579322815, "brier": 0.045359522104263306, "ece": 0.05632979728086318, "accuracy": 0.9463504552841187 }, "beta": { "mean": 0.3987804353237152, "min": 0.021586544811725616, "max": 0.6299230456352234 }, "bias": { "mean": -0.60589200258255, "min": -1.0286215543746948, "max": 0.5900660753250122 }, "groups": { "overconfident": { "texts": 810, "uncalibrated": { "nll": 0.396918922662735, "brier": 0.11602238565683365, "ece": 0.1158942031665317, "accuracy": 0.8361965417861938 }, "global_scaling": { "nll": 0.2803961932659149, "brier": 0.09336278587579727, "ece": 0.07126283364916436, "accuracy": 0.8470292091369629 }, "calibrated": { "nll": 0.22382210195064545, "brier": 0.07280737161636353, "ece": 0.0505468512780157, "accuracy": 0.8692417144775391 } }, "underconfident": { "texts": 952, "uncalibrated": { "nll": 0.18349525332450867, "brier": 0.04817432537674904, "ece": 0.0770668854550236, "accuracy": 0.9513456225395203 }, "global_scaling": { "nll": 0.21706710755825043, "brier": 0.055873360484838486, "ece": 0.09890477648015017, "accuracy": 0.9416147470474243 }, "calibrated": { "nll": 0.24962961673736572, "brier": 0.06872871518135071, "ece": 0.07814065774855061, "accuracy": 0.9042116403579712 } }, "mixed": { "texts": 335, "uncalibrated": { "nll": 0.2900349199771881, "brier": 0.0966075211763382, "ece": 0.08953729352259154, "accuracy": 0.8737332224845886 }, "global_scaling": { "nll": 0.25439244508743286, "brier": 0.08061423152685165, "ece": 0.08066619354176002, "accuracy": 0.8430566787719727 }, "calibrated": { "nll": 0.2466566115617752, "brier": 0.07597621530294418, "ece": 0.049607173441170604, "accuracy": 0.8896192908287048 } }, "normal": { "texts": 5655, "uncalibrated": { "nll": 0.03119591809809208, "brier": 0.006655354518443346, "ece": 0.024245568390645844, "accuracy": 0.985680341720581 }, "global_scaling": { "nll": 0.05599905923008919, "brier": 0.010298365727066994, "ece": 0.042321458578928245, "accuracy": 0.9956013560295105 }, "calibrated": { "nll": 0.1182105615735054, "brier": 0.028781907632946968, "ece": 0.07998285826653084, "accuracy": 0.9845244884490967 } } }, "softmax": { "texts": 6756, "uncalibrated": { "nll": 0.60243159532547, "brier": 0.25821787118911743, "ece": 0.08630393553886592, "accuracy": 0.8288928270339966 }, "global_scaling": { "nll": 0.47443029284477234, "brier": 0.23971512913703918, "ece": 0.035015895881971036, "accuracy": 0.8288928270339966 }, "calibrated": { "nll": 0.5013883709907532, "brier": 0.25799572467803955, "ece": 0.08973347695110602, "accuracy": 0.8288928270339966 } } }, "held_out": { "uncalibrated": { "nll": 0.134813591837883, "brier": 0.0394577793776989, "ece": 0.009385065096777556, "accuracy": 0.9442779421806335 }, "global_scaling": { "nll": 0.1316956728696823, "brier": 0.0370788536965847, "ece": 0.024935297782987448, "accuracy": 0.9489754438400269 }, "calibrated": { "nll": 0.1662309616804123, "brier": 0.046702150255441666, "ece": 0.05650961722975578, "accuracy": 0.9439539909362793 }, "beta": { "mean": 0.39366212487220764, "min": 0.023016568273305893, "max": 0.6071414947509766 }, "bias": { "mean": -0.5795870423316956, "min": -1.0330911874771118, "max": 0.5484000444412231 }, "groups": { "overconfident": { "texts": 180, "uncalibrated": { "nll": 0.5719922184944153, "brier": 0.17103661596775055, "ece": 0.17139117058701653, "accuracy": 0.7580760717391968 }, "global_scaling": { "nll": 0.401309609413147, "brier": 0.1368330717086792, "ece": 0.11016014898095755, "accuracy": 0.7717157006263733 }, "calibrated": { "nll": 0.31307777762413025, "brier": 0.10509755462408066, "ece": 0.07207305132336525, "accuracy": 0.8083273768424988 } }, "underconfident": { "texts": 224, "uncalibrated": { "nll": 0.19197306036949158, "brier": 0.05295313894748688, "ece": 0.06992532354870229, "accuracy": 0.9405063390731812 }, "global_scaling": { "nll": 0.2224033921957016, "brier": 0.0592648983001709, "ece": 0.09383183797541672, "accuracy": 0.9335442781448364 }, "calibrated": { "nll": 0.2595052719116211, "brier": 0.07272262871265411, "ece": 0.07825523673930124, "accuracy": 0.8962025046348572 } }, "mixed": { "texts": 97, "uncalibrated": { "nll": 0.29701313376426697, "brier": 0.10088907927274704, "ece": 0.10026825751042845, "accuracy": 0.8682310581207275 }, "global_scaling": { "nll": 0.2555745542049408, "brier": 0.08258884400129318, "ece": 0.08199416873906554, "accuracy": 0.8375450968742371 }, "calibrated": { "nll": 0.255821168422699, "brier": 0.08022957295179367, "ece": 0.04236628423505007, "accuracy": 0.8835740089416504 } }, "normal": { "texts": 1436, "uncalibrated": { "nll": 0.028472088277339935, "brier": 0.006469887215644121, "ece": 0.022601176297401007, "accuracy": 0.9865714907646179 }, "global_scaling": { "nll": 0.052316512912511826, "brier": 0.009927067905664444, "ece": 0.04092645116679347, "accuracy": 0.9967336058616638 }, "calibrated": { "nll": 0.11164623498916626, "brier": 0.02739347331225872, "ece": 0.0744886720722503, "accuracy": 0.9840309619903564 } } }, "softmax": { "texts": 1709, "uncalibrated": { "nll": 0.5652890205383301, "brier": 0.25095003843307495, "ece": 0.08337326918689637, "accuracy": 0.8338209390640259 }, "global_scaling": { "nll": 0.4503483176231384, "brier": 0.23204359412193298, "ece": 0.036847078234715565, "accuracy": 0.8338209390640259 }, "calibrated": { "nll": 0.4841042459011078, "brier": 0.25139153003692627, "ece": 0.08952697261251302, "accuracy": 0.8338209390640259 } } }, "args": { "model_path": "models/qwen3.8-9b-fast/last", "data_path": "data/decision_index_calibration.json", "save_path": "models/qwen3.8-9b-fast/last-calibrated", "eval_ratio": 0.2, "max_examples": 10000, "max_length": 2048, "batch_size": 8, "num_workers": 4, "dtype": "bfloat16", "hidden_size": 256, "beta_max": 3.0, "dropout": 0.1, "prior_weight": 0.01, "use_bias": true, "bias_prior_weight": 0.01, "skip_long": true, "rank_preserving": true, "early_stopping": null, "objective": "mixed", "softmax_weight": 0.5, "balance_confidence": true, "confidence_threshold": 0.7, "num_epochs": 3, "calibrator_batch_size": 64, "lr": 0.0001, "weight_decay": 0.01, "patience": 3, "seed": 42 } }