moderation / eval /calibration.json
bogdanraduta's picture
Upload eval/calibration.json with huggingface_hub
9acf05e verified
Raw History Blame Contribute Delete
12.8 kB
{
"threshold": 0.91,
"objective": "macro_f1",
"value_on_val": 0.9037,
"default_threshold": 0.5,
"default_value_on_val": 0.8658,
"improvement": 0.0379,
"chosen_on": "validation split, never test",
"curve": [
{
"threshold": 0.02,
"macro_f1": 0.1503,
"any_label_recall": 1.0,
"any_label_fpr": 0.1051
},
{
"threshold": 0.03,
"macro_f1": 0.1631,
"any_label_recall": 1.0,
"any_label_fpr": 0.0813
},
{
"threshold": 0.04,
"macro_f1": 0.2075,
"any_label_recall": 1.0,
"any_label_fpr": 0.0747
},
{
"threshold": 0.05,
"macro_f1": 0.2941,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0701
},
{
"threshold": 0.06,
"macro_f1": 0.3936,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0628
},
{
"threshold": 0.07,
"macro_f1": 0.4777,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0588
},
{
"threshold": 0.08,
"macro_f1": 0.5395,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0582
},
{
"threshold": 0.09,
"macro_f1": 0.5885,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0562
},
{
"threshold": 0.1,
"macro_f1": 0.6306,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0549
},
{
"threshold": 0.11,
"macro_f1": 0.6592,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0549
},
{
"threshold": 0.12,
"macro_f1": 0.6812,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0535
},
{
"threshold": 0.13,
"macro_f1": 0.7026,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0529
},
{
"threshold": 0.14,
"macro_f1": 0.7188,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0516
},
{
"threshold": 0.15,
"macro_f1": 0.7321,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0516
},
{
"threshold": 0.16,
"macro_f1": 0.7422,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0502
},
{
"threshold": 0.17,
"macro_f1": 0.7524,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0489
},
{
"threshold": 0.18,
"macro_f1": 0.7614,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0489
},
{
"threshold": 0.19,
"macro_f1": 0.7698,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0476
},
{
"threshold": 0.2,
"macro_f1": 0.7787,
"any_label_recall": 0.9993,
"any_label_fpr": 0.0469
},
{
"threshold": 0.21,
"macro_f1": 0.7856,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0463
},
{
"threshold": 0.22,
"macro_f1": 0.7921,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0463
},
{
"threshold": 0.23,
"macro_f1": 0.7957,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0463
},
{
"threshold": 0.24,
"macro_f1": 0.8001,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0456
},
{
"threshold": 0.25,
"macro_f1": 0.8041,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0456
},
{
"threshold": 0.26,
"macro_f1": 0.8076,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0449
},
{
"threshold": 0.27,
"macro_f1": 0.8114,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0443
},
{
"threshold": 0.28,
"macro_f1": 0.8158,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0443
},
{
"threshold": 0.29,
"macro_f1": 0.8198,
"any_label_recall": 0.9987,
"any_label_fpr": 0.043
},
{
"threshold": 0.3,
"macro_f1": 0.8237,
"any_label_recall": 0.9987,
"any_label_fpr": 0.043
},
{
"threshold": 0.31,
"macro_f1": 0.8281,
"any_label_recall": 0.9987,
"any_label_fpr": 0.043
},
{
"threshold": 0.32,
"macro_f1": 0.8308,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0423
},
{
"threshold": 0.33,
"macro_f1": 0.8345,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0423
},
{
"threshold": 0.34,
"macro_f1": 0.836,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0416
},
{
"threshold": 0.35,
"macro_f1": 0.8386,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0416
},
{
"threshold": 0.36,
"macro_f1": 0.8408,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0416
},
{
"threshold": 0.37,
"macro_f1": 0.8434,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0403
},
{
"threshold": 0.38,
"macro_f1": 0.846,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0403
},
{
"threshold": 0.39,
"macro_f1": 0.849,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0403
},
{
"threshold": 0.4,
"macro_f1": 0.8504,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0397
},
{
"threshold": 0.41,
"macro_f1": 0.8522,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0383
},
{
"threshold": 0.42,
"macro_f1": 0.8547,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0377
},
{
"threshold": 0.43,
"macro_f1": 0.856,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0377
},
{
"threshold": 0.44,
"macro_f1": 0.8583,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0377
},
{
"threshold": 0.45,
"macro_f1": 0.8603,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0377
},
{
"threshold": 0.46,
"macro_f1": 0.8623,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0377
},
{
"threshold": 0.47,
"macro_f1": 0.8636,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0377
},
{
"threshold": 0.48,
"macro_f1": 0.8648,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0377
},
{
"threshold": 0.49,
"macro_f1": 0.8662,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0377
},
{
"threshold": 0.5,
"macro_f1": 0.8658,
"any_label_recall": 0.9987,
"any_label_fpr": 0.037
},
{
"threshold": 0.51,
"macro_f1": 0.8676,
"any_label_recall": 0.9987,
"any_label_fpr": 0.037
},
{
"threshold": 0.52,
"macro_f1": 0.8699,
"any_label_recall": 0.9987,
"any_label_fpr": 0.037
},
{
"threshold": 0.53,
"macro_f1": 0.8708,
"any_label_recall": 0.9987,
"any_label_fpr": 0.037
},
{
"threshold": 0.54,
"macro_f1": 0.8738,
"any_label_recall": 0.9987,
"any_label_fpr": 0.037
},
{
"threshold": 0.55,
"macro_f1": 0.8748,
"any_label_recall": 0.9987,
"any_label_fpr": 0.037
},
{
"threshold": 0.56,
"macro_f1": 0.877,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0364
},
{
"threshold": 0.57,
"macro_f1": 0.8779,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0364
},
{
"threshold": 0.58,
"macro_f1": 0.8785,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0357
},
{
"threshold": 0.59,
"macro_f1": 0.8805,
"any_label_recall": 0.9987,
"any_label_fpr": 0.035
},
{
"threshold": 0.6,
"macro_f1": 0.882,
"any_label_recall": 0.9987,
"any_label_fpr": 0.0337
},
{
"threshold": 0.61,
"macro_f1": 0.8823,
"any_label_recall": 0.998,
"any_label_fpr": 0.033
},
{
"threshold": 0.62,
"macro_f1": 0.8833,
"any_label_recall": 0.998,
"any_label_fpr": 0.033
},
{
"threshold": 0.63,
"macro_f1": 0.8837,
"any_label_recall": 0.9974,
"any_label_fpr": 0.033
},
{
"threshold": 0.64,
"macro_f1": 0.8845,
"any_label_recall": 0.9974,
"any_label_fpr": 0.033
},
{
"threshold": 0.65,
"macro_f1": 0.8838,
"any_label_recall": 0.9974,
"any_label_fpr": 0.033
},
{
"threshold": 0.66,
"macro_f1": 0.8844,
"any_label_recall": 0.9974,
"any_label_fpr": 0.033
},
{
"threshold": 0.67,
"macro_f1": 0.8853,
"any_label_recall": 0.9974,
"any_label_fpr": 0.0324
},
{
"threshold": 0.68,
"macro_f1": 0.8877,
"any_label_recall": 0.9974,
"any_label_fpr": 0.0317
},
{
"threshold": 0.69,
"macro_f1": 0.8889,
"any_label_recall": 0.9974,
"any_label_fpr": 0.0304
},
{
"threshold": 0.7,
"macro_f1": 0.8894,
"any_label_recall": 0.9974,
"any_label_fpr": 0.0304
},
{
"threshold": 0.71,
"macro_f1": 0.8901,
"any_label_recall": 0.9974,
"any_label_fpr": 0.0297
},
{
"threshold": 0.72,
"macro_f1": 0.8908,
"any_label_recall": 0.9974,
"any_label_fpr": 0.0291
},
{
"threshold": 0.73,
"macro_f1": 0.8924,
"any_label_recall": 0.9974,
"any_label_fpr": 0.0291
},
{
"threshold": 0.74,
"macro_f1": 0.8931,
"any_label_recall": 0.9974,
"any_label_fpr": 0.0291
},
{
"threshold": 0.75,
"macro_f1": 0.8943,
"any_label_recall": 0.9974,
"any_label_fpr": 0.0284
},
{
"threshold": 0.76,
"macro_f1": 0.8962,
"any_label_recall": 0.9974,
"any_label_fpr": 0.0284
},
{
"threshold": 0.77,
"macro_f1": 0.8973,
"any_label_recall": 0.9967,
"any_label_fpr": 0.0278
},
{
"threshold": 0.78,
"macro_f1": 0.8977,
"any_label_recall": 0.9961,
"any_label_fpr": 0.0278
},
{
"threshold": 0.79,
"macro_f1": 0.8986,
"any_label_recall": 0.9954,
"any_label_fpr": 0.0271
},
{
"threshold": 0.8,
"macro_f1": 0.8999,
"any_label_recall": 0.9954,
"any_label_fpr": 0.0271
},
{
"threshold": 0.81,
"macro_f1": 0.8997,
"any_label_recall": 0.9948,
"any_label_fpr": 0.0271
},
{
"threshold": 0.82,
"macro_f1": 0.8996,
"any_label_recall": 0.9935,
"any_label_fpr": 0.0264
},
{
"threshold": 0.83,
"macro_f1": 0.8996,
"any_label_recall": 0.9928,
"any_label_fpr": 0.0258
},
{
"threshold": 0.84,
"macro_f1": 0.8996,
"any_label_recall": 0.9915,
"any_label_fpr": 0.0258
},
{
"threshold": 0.85,
"macro_f1": 0.8989,
"any_label_recall": 0.9902,
"any_label_fpr": 0.0251
},
{
"threshold": 0.86,
"macro_f1": 0.8995,
"any_label_recall": 0.9863,
"any_label_fpr": 0.0251
},
{
"threshold": 0.87,
"macro_f1": 0.9,
"any_label_recall": 0.9831,
"any_label_fpr": 0.0245
},
{
"threshold": 0.88,
"macro_f1": 0.9004,
"any_label_recall": 0.9804,
"any_label_fpr": 0.0245
},
{
"threshold": 0.89,
"macro_f1": 0.9018,
"any_label_recall": 0.9785,
"any_label_fpr": 0.0231
},
{
"threshold": 0.9,
"macro_f1": 0.9032,
"any_label_recall": 0.9759,
"any_label_fpr": 0.0212
},
{
"threshold": 0.91,
"macro_f1": 0.9037,
"any_label_recall": 0.9687,
"any_label_fpr": 0.0185
},
{
"threshold": 0.92,
"macro_f1": 0.9027,
"any_label_recall": 0.9583,
"any_label_fpr": 0.0165
},
{
"threshold": 0.93,
"macro_f1": 0.8994,
"any_label_recall": 0.9472,
"any_label_fpr": 0.0152
},
{
"threshold": 0.94,
"macro_f1": 0.8972,
"any_label_recall": 0.9329,
"any_label_fpr": 0.0126
},
{
"threshold": 0.95,
"macro_f1": 0.8979,
"any_label_recall": 0.914,
"any_label_fpr": 0.0106
}
],
"notes": [],
"note": "**This threshold is not a tuned parameter, and the shipped policy default stays at 0.84.** Three runs against this corpus, two seed replicates and this retrain, read 0.69, 0.83 and 0.81, and the validation curve is flat across the whole range that matters: macro F1 moves from 0.9288 at 0.50 to a peak of 0.9372 at 0.81, a spread of 0.0084 over 45 points of threshold. 0.84 reads 0.9369 on the same curve, indistinguishable from the calibrated value. 0.84 is the reviewed value shipped before this retrain and is unchanged by it."
}