{ "objective": -0.5460977168843815, "paired_flip": { "pairs": 80, "flip_rate": 0.55, "both_correct_rate": 0.45, "invariant_pairs": 32, "invariance_rate": 0.9375, "invariant_both_correct_rate": 0.46875 }, "unknowable": null, "clean": { "n": 1264, "nll": 0.5215492407482734, "acc": 0.7943037974683544, "ece": 0.04048827875950509, "brier": 0.27717776074983425, "mean_conf": 0.8284767038904833, "confident_error_rate": 0.028481012658227847, "coverage_at_0_9": 0.5561708860759493, "accuracy_at_0_9": 0.9487908961593172, "coverage_at_5pct_error": 0.555379746835443, "coverage_at_1pct_error": 0.006329113924050633, "aurc": 0.07067902977682425, "error_rate_at_0_9": 0.051209103840682786, "confidence_bias": 0.03417290642212889, "top_bins": { "0.9": { "n": 703, "errors": 36, "error_rate": 0.051209103840682786 }, "0.95": { "n": 646, "errors": 27, "error_rate": 0.04179566563467492 }, "0.99": { "n": 512, "errors": 13, "error_rate": 0.025390625 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.9588607594936709, "confidence_cutoff": 0.961561151958084 }, "0.8": { "coverage": 0.8006329113924051, "accuracy": 0.866600790513834, "confidence_cutoff": 0.5841912218049162 } }, "score_mae": 0.4606092452508017, "ranked_probability_score": 0.07134033027857928 }, "tasks": { "agnews": { "n": 80, "nll": 0.7253619615276804, "acc": 0.825, "ece": 0.12435274294907851, "brier": 0.29748699997093375, "mean_conf": 0.9467044212167703, "confident_error_rate": 0.1125, "coverage_at_0_9": 0.8375, "accuracy_at_0_9": 0.8656716417910447, "coverage_at_5pct_error": 0.5, "coverage_at_1pct_error": 0.025, "aurc": 0.0891676305179602, "error_rate_at_0_9": 0.13432835820895522, "confidence_bias": 0.12170442121677039, "top_bins": { "0.9": { "n": 67, "errors": 9, "error_rate": 0.13432835820895522 }, "0.95": { "n": 60, "errors": 5, "error_rate": 0.08333333333333333 }, "0.99": { "n": 47, "errors": 3, "error_rate": 0.06382978723404255 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.95, "confidence_cutoff": 0.9968383335724911 }, "0.8": { "coverage": 0.8, "accuracy": 0.875, "confidence_cutoff": 0.9297638903048708 } } }, "agnews_yn": { "n": 160, "nll": 0.27274937177459047, "acc": 0.8875, "ece": 0.07205353407763021, "brier": 0.16638284259750016, "mean_conf": 0.9492791941714842, "confident_error_rate": 0.025, "coverage_at_0_9": 0.81875, "accuracy_at_0_9": 0.9694656488549618, "coverage_at_5pct_error": 0.8375, "coverage_at_1pct_error": 0.625, "aurc": 0.019553552474504545, "error_rate_at_0_9": 0.030534351145038167, "confidence_bias": 0.06177919417148425, "top_bins": { "0.9": { "n": 131, "errors": 4, "error_rate": 0.030534351145038167 }, "0.95": { "n": 122, "errors": 3, "error_rate": 0.02459016393442623 }, "0.99": { "n": 97, "errors": 1, "error_rate": 0.010309278350515464 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.9946307004104364 }, "0.8": { "coverage": 0.8, "accuracy": 0.9765625, "confidence_cutoff": 0.9334476480940993 } } }, "amazon": { "n": 80, "nll": 0.9079064857116548, "acc": 0.675, "ece": 0.09082075584790608, "brier": 0.48731382188074246, "mean_conf": 0.6600554720427965, "confident_error_rate": 0.0, "coverage_at_0_9": 0.1125, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 0.25, "coverage_at_1pct_error": 0.2375, "aurc": 0.19069566998937534, "error_rate_at_0_9": 0.0, "confidence_bias": -0.01494452795720358, "top_bins": { "0.9": { "n": 9, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 3, "errors": 0, "error_rate": 0.0 }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.775, "confidence_cutoff": 0.6450436740808796 }, "0.8": { "coverage": 0.8, "accuracy": 0.71875, "confidence_cutoff": 0.48814962988854316 } }, "score_mae": 0.4846846425037448, "ranked_probability_score": 0.07588150660688607 }, "banking77": { "n": 80, "nll": 0.6355525688857636, "acc": 0.8375, "ece": 0.07048704096779947, "brier": 0.24473580199909098, "mean_conf": 0.8527902873638666, "confident_error_rate": 0.025, "coverage_at_0_9": 0.65, "accuracy_at_0_9": 0.9615384615384616, "coverage_at_5pct_error": 0.65, "coverage_at_1pct_error": 0.5625, "aurc": 0.03531804637364592, "error_rate_at_0_9": 0.038461538461538464, "confidence_bias": 0.015290287363866617, "top_bins": { "0.9": { "n": 52, "errors": 2, "error_rate": 0.038461538461538464 }, "0.95": { "n": 47, "errors": 1, "error_rate": 0.02127659574468085 }, "0.99": { "n": 32, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.9819397006054039 }, "0.8": { "coverage": 0.8, "accuracy": 0.921875, "confidence_cutoff": 0.64788273781478 } } }, "boolq": { "n": 80, "nll": 0.5330866725434655, "acc": 0.7875, "ece": 0.11078594898530592, "brier": 0.32181604785225043, "mean_conf": 0.8668040159586322, "confident_error_rate": 0.0625, "coverage_at_0_9": 0.5875, "accuracy_at_0_9": 0.8936170212765957, "coverage_at_5pct_error": 0.35, "coverage_at_1pct_error": 0.15, "aurc": 0.10319650984358723, "error_rate_at_0_9": 0.10638297872340426, "confidence_bias": 0.07930401595863223, "top_bins": { "0.9": { "n": 47, "errors": 5, "error_rate": 0.10638297872340426 }, "0.95": { "n": 40, "errors": 4, "error_rate": 0.1 }, "0.99": { "n": 17, "errors": 1, "error_rate": 0.058823529411764705 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.9, "confidence_cutoff": 0.960176618093949 }, "0.8": { "coverage": 0.8, "accuracy": 0.8125, "confidence_cutoff": 0.7061129387549795 } } }, "composition_atom": { "n": 16, "nll": 0.6868630202544777, "acc": 0.5625, "ece": 0.04684587795814621, "brier": 0.49376454489058674, "mean_conf": 0.5156541220418538, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.125, "coverage_at_1pct_error": 0.125, "aurc": 0.40027177683427684, "error_rate_at_0_9": null, "confidence_bias": -0.04684587795814621, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5625, "accuracy": 0.5555555555555556, "confidence_cutoff": 0.5104723876996271 }, "0.8": { "coverage": 0.8125, "accuracy": 0.6153846153846154, "confidence_cutoff": 0.5015301704406738 } } }, "composition_conditional": { "n": 16, "nll": 0.692639150220755, "acc": 0.625, "ece": 0.11956254686423151, "brier": 0.4994918868193602, "mean_conf": 0.5054374531357685, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.1875, "coverage_at_1pct_error": 0.1875, "aurc": 0.36212017149517156, "error_rate_at_0_9": null, "confidence_bias": -0.11956254686423151, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.375, "confidence_cutoff": 0.5062627045433907 }, "0.8": { "coverage": 0.8125, "accuracy": 0.6153846153846154, "confidence_cutoff": 0.5018942356109619 } } }, "composition_conjunction": { "n": 16, "nll": 0.6807725943312114, "acc": 0.75, "ece": 0.24073381449590237, "brier": 0.4876279322120418, "mean_conf": 0.5092661855040976, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.5625, "coverage_at_1pct_error": 0.5625, "aurc": 0.07196657509157509, "error_rate_at_0_9": null, "confidence_bias": -0.24073381449590237, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.5111942734958448 }, "0.8": { "coverage": 0.8125, "accuracy": 0.8461538461538461, "confidence_cutoff": 0.506822407245636 } } }, "composition_disjunction": { "n": 16, "nll": 0.6845496087519543, "acc": 0.625, "ece": 0.10940085910388908, "brier": 0.4914097560792531, "mean_conf": 0.5155991408961109, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.125, "coverage_at_1pct_error": 0.125, "aurc": 0.3235514485514486, "error_rate_at_0_9": null, "confidence_bias": -0.10940085910388908, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.5, "confidence_cutoff": 0.5133245877430088 }, "0.8": { "coverage": 0.8125, "accuracy": 0.6153846153846154, "confidence_cutoff": 0.5098766532577081 } } }, "composition_exception": { "n": 16, "nll": 0.6999479317489916, "acc": 0.3125, "ece": 0.19967369551773084, "brier": 0.5067955467438511, "mean_conf": 0.5121736955177308, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.0625, "coverage_at_1pct_error": 0.0625, "aurc": 0.6147534756909757, "error_rate_at_0_9": null, "confidence_bias": 0.19967369551773084, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.25, "confidence_cutoff": 0.5091820359230042 }, "0.8": { "coverage": 0.8125, "accuracy": 0.38461538461538464, "confidence_cutoff": 0.5007553100585938 } } }, "composition_negation": { "n": 16, "nll": 0.7255260889583809, "acc": 0.3125, "ece": 0.21004319515882908, "brier": 0.5322663751971064, "mean_conf": 0.5225431951588291, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.0, "coverage_at_1pct_error": 0.0, "aurc": 0.8313863220113219, "error_rate_at_0_9": null, "confidence_bias": 0.21004319515882908, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.125, "confidence_cutoff": 0.5219886458512737 }, "0.8": { "coverage": 0.8125, "accuracy": 0.38461538461538464, "confidence_cutoff": 0.5036663563152062 } } }, "composition_nested_and": { "n": 16, "nll": 0.6977710170099051, "acc": 0.4375, "ece": 0.07430293594870152, "brier": 0.5046165383435348, "mean_conf": 0.5118029359487015, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.0, "coverage_at_1pct_error": 0.0, "aurc": 0.6242112921800421, "error_rate_at_0_9": null, "confidence_bias": 0.07430293594870152, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5625, "accuracy": 0.4444444444444444, "confidence_cutoff": 0.5065592376423109 }, "0.8": { "coverage": 0.8125, "accuracy": 0.5384615384615384, "confidence_cutoff": 0.5032630862750462 } } }, "composition_nested_or": { "n": 16, "nll": 0.6933208400499358, "acc": 0.375, "ece": 0.13961227642659346, "brier": 0.5001786152290317, "mean_conf": 0.5146122764265935, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.1875, "coverage_at_1pct_error": 0.1875, "aurc": 0.4065724206349206, "error_rate_at_0_9": null, "confidence_bias": 0.13961227642659346, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.5, "confidence_cutoff": 0.5179873704910278 }, "0.8": { "coverage": 0.875, "accuracy": 0.42857142857142855, "confidence_cutoff": 0.5046656131744385 } } }, "contrastive_age_eligibility": { "n": 24, "nll": 0.0019957258438999047, "acc": 1.0, "ece": 0.001986453041521785, "brier": 3.665966146860536e-05, "mean_conf": 0.9980135469584782, "confident_error_rate": 0.0, "coverage_at_0_9": 1.0, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 1.0, "coverage_at_1pct_error": 1.0, "aurc": 0.0, "error_rate_at_0_9": 0.0, "confidence_bias": -0.001986453041521785, "top_bins": { "0.9": { "n": 24, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 24, "errors": 0, "error_rate": 0.0 }, "0.99": { "n": 23, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.9991325300448652 }, "0.8": { "coverage": 0.8333333333333334, "accuracy": 1.0, "confidence_cutoff": 0.9977411031723022 } } }, "contrastive_quantity_limit": { "n": 24, "nll": 0.2953036070445502, "acc": 0.875, "ece": 0.08720234763968863, "brier": 0.19980341420601166, "mean_conf": 0.7877976523603114, "confident_error_rate": 0.0, "coverage_at_0_9": 0.5, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 0.8333333333333334, "coverage_at_1pct_error": 0.75, "aurc": 0.020863970284736872, "error_rate_at_0_9": 0.0, "confidence_bias": -0.0872023476396886, "top_bins": { "0.9": { "n": 12, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 12, "errors": 0, "error_rate": 0.0 }, "0.99": { "n": 12, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.999568032339159 }, "0.8": { "coverage": 0.8333333333333334, "accuracy": 0.95, "confidence_cutoff": 0.5362308549929372 } } }, "contrastive_return_window": { "n": 24, "nll": 0.6873419627659144, "acc": 0.4166666666666667, "ece": 0.19777798255950627, "brier": 0.4963496570710506, "mean_conf": 0.6110676657072444, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.125, "coverage_at_1pct_error": 0.125, "aurc": 0.3518959178145307, "error_rate_at_0_9": null, "confidence_bias": 0.19440099904057767, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.5833333333333334, "confidence_cutoff": 0.606834017823099 }, "0.8": { "coverage": 0.8333333333333334, "accuracy": 0.5, "confidence_cutoff": 0.5422527466861756 } } }, "contrastive_spend_threshold": { "n": 24, "nll": 0.00036495022640064607, "acc": 1.0, "ece": 0.00036484651323320794, "brier": 4.018898659036875e-07, "mean_conf": 0.9996351534867668, "confident_error_rate": 0.0, "coverage_at_0_9": 1.0, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 1.0, "coverage_at_1pct_error": 1.0, "aurc": 0.0, "error_rate_at_0_9": 0.0, "confidence_bias": -0.00036484651323320794, "top_bins": { "0.9": { "n": 24, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 24, "errors": 0, "error_rate": 0.0 }, "0.99": { "n": 24, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.9998236097452727 }, "0.8": { "coverage": 0.8333333333333334, "accuracy": 1.0, "confidence_cutoff": 0.9994718056567214 } } }, "dbpedia14": { "n": 80, "nll": 0.421653557685442, "acc": 0.9625, "ece": 0.038590716499222494, "brier": 0.0749984906704935, "mean_conf": 0.9977291409078489, "confident_error_rate": 0.0375, "coverage_at_0_9": 0.9875, "accuracy_at_0_9": 0.9620253164556962, "coverage_at_5pct_error": 1.0, "coverage_at_1pct_error": 0.1, "aurc": 0.029371556564368212, "error_rate_at_0_9": 0.0379746835443038, "confidence_bias": 0.0352291409078489, "top_bins": { "0.9": { "n": 79, "errors": 3, "error_rate": 0.0379746835443038 }, "0.95": { "n": 79, "errors": 3, "error_rate": 0.0379746835443038 }, "0.99": { "n": 78, "errors": 3, "error_rate": 0.038461538461538464 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.975, "confidence_cutoff": 0.9999933016730673 }, "0.8": { "coverage": 0.8, "accuracy": 0.984375, "confidence_cutoff": 0.9999585960145762 } } }, "imdb": { "n": 80, "nll": 0.33141699422220733, "acc": 0.9125, "ece": 0.05920801680085907, "brier": 0.14056258642338354, "mean_conf": 0.9717080168008593, "confident_error_rate": 0.05, "coverage_at_0_9": 0.925, "accuracy_at_0_9": 0.9459459459459459, "coverage_at_5pct_error": 0.85, "coverage_at_1pct_error": 0.05, "aurc": 0.046590820152253506, "error_rate_at_0_9": 0.05405405405405406, "confidence_bias": 0.0592080168008593, "top_bins": { "0.9": { "n": 74, "errors": 4, "error_rate": 0.05405405405405406 }, "0.95": { "n": 73, "errors": 4, "error_rate": 0.0547945205479452 }, "0.99": { "n": 61, "errors": 2, "error_rate": 0.03278688524590164 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.975, "confidence_cutoff": 0.9990444188595242 }, "0.8": { "coverage": 0.8, "accuracy": 0.953125, "confidence_cutoff": 0.9856349119042156 } } }, "mnli": { "n": 80, "nll": 0.6238560572386006, "acc": 0.775, "ece": 0.098545258596266, "brier": 0.33638415666106714, "mean_conf": 0.8037648115227398, "confident_error_rate": 0.05, "coverage_at_0_9": 0.3625, "accuracy_at_0_9": 0.8620689655172413, "coverage_at_5pct_error": 0.25, "coverage_at_1pct_error": 0.025, "aurc": 0.1461657498448564, "error_rate_at_0_9": 0.13793103448275862, "confidence_bias": 0.028764811522739797, "top_bins": { "0.9": { "n": 29, "errors": 4, "error_rate": 0.13793103448275862 }, "0.95": { "n": 23, "errors": 2, "error_rate": 0.08695652173913043 }, "0.99": { "n": 10, "errors": 1, "error_rate": 0.1 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.85, "confidence_cutoff": 0.8368562883553604 }, "0.8": { "coverage": 0.8, "accuracy": 0.8125, "confidence_cutoff": 0.6399822597195601 } } }, "sst5": { "n": 80, "nll": 0.9312577104784421, "acc": 0.6375, "ece": 0.058047224154593555, "brier": 0.517886063473853, "mean_conf": 0.6155482010149003, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.0, "coverage_at_1pct_error": 0.0, "aurc": 0.28378138053885427, "error_rate_at_0_9": null, "confidence_bias": -0.02195179898509969, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.7, "confidence_cutoff": 0.6130724488092862 }, "0.8": { "coverage": 0.8, "accuracy": 0.671875, "confidence_cutoff": 0.4991830745359043 } }, "score_mae": 0.5024858053525012, "ranked_probability_score": 0.07990343060681752 }, "trec": { "n": 80, "nll": 0.31463663041157075, "acc": 0.95, "ece": 0.05342332897478768, "brier": 0.10129446206548179, "mean_conf": 0.9916042689599147, "confident_error_rate": 0.05, "coverage_at_0_9": 0.975, "accuracy_at_0_9": 0.9487179487179487, "coverage_at_5pct_error": 1.0, "coverage_at_1pct_error": 0.5625, "aurc": 0.012206716405033919, "error_rate_at_0_9": 0.05128205128205128, "confidence_bias": 0.04160426895991476, "top_bins": { "0.9": { "n": 78, "errors": 4, "error_rate": 0.05128205128205128 }, "0.95": { "n": 78, "errors": 4, "error_rate": 0.05128205128205128 }, "0.99": { "n": 72, "errors": 2, "error_rate": 0.027777777777777776 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.9994122098980961 }, "0.8": { "coverage": 0.8, "accuracy": 0.96875, "confidence_cutoff": 0.9962499645174014 } } }, "yelp": { "n": 80, "nll": 0.7093590738165945, "acc": 0.675, "ece": 0.09823356777966172, "brier": 0.4286367817783625, "mean_conf": 0.7157254348808417, "confident_error_rate": 0.0, "coverage_at_0_9": 0.1625, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 0.2, "coverage_at_1pct_error": 0.2, "aurc": 0.17643766746740774, "error_rate_at_0_9": 0.0, "confidence_bias": 0.04072543488084168, "top_bins": { "0.9": { "n": 13, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 8, "errors": 0, "error_rate": 0.0 }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.825, "confidence_cutoff": 0.7173906410209289 }, "0.8": { "coverage": 0.8, "accuracy": 0.71875, "confidence_cutoff": 0.5784675950234985 } }, "score_mae": 0.39465728789615917, "ranked_probability_score": 0.058236053622034226 }, "yelp_yn": { "n": 80, "nll": 0.15311162372276566, "acc": 0.95, "ece": 0.04194291267597787, "brier": 0.08344044292524996, "mean_conf": 0.9285674559777203, "confident_error_rate": 0.0125, "coverage_at_0_9": 0.8, "accuracy_at_0_9": 0.984375, "coverage_at_5pct_error": 1.0, "coverage_at_1pct_error": 0.5625, "aurc": 0.010052232246343523, "error_rate_at_0_9": 0.015625, "confidence_bias": -0.021432544022279676, "top_bins": { "0.9": { "n": 64, "errors": 1, "error_rate": 0.015625 }, "0.95": { "n": 53, "errors": 1, "error_rate": 0.018867924528301886 }, "0.99": { "n": 39, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.9892096070827485 }, "0.8": { "coverage": 0.8, "accuracy": 0.984375, "confidence_cutoff": 0.9120315109484305 } } } }, "variants": { "clean": { "n": 1264, "nll": 0.5215492407482734, "acc": 0.7943037974683544, "ece": 0.04048827875950509, "brier": 0.27717776074983425, "mean_conf": 0.8284767038904833, "confident_error_rate": 0.028481012658227847, "coverage_at_0_9": 0.5561708860759493, "accuracy_at_0_9": 0.9487908961593172, "coverage_at_5pct_error": 0.555379746835443, "coverage_at_1pct_error": 0.006329113924050633, "aurc": 0.07067902977682425, "error_rate_at_0_9": 0.051209103840682786, "confidence_bias": 0.03417290642212889, "top_bins": { "0.9": { "n": 703, "errors": 36, "error_rate": 0.051209103840682786 }, "0.95": { "n": 646, "errors": 27, "error_rate": 0.04179566563467492 }, "0.99": { "n": 512, "errors": 13, "error_rate": 0.025390625 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.9588607594936709, "confidence_cutoff": 0.961561151958084 }, "0.8": { "coverage": 0.8006329113924051, "accuracy": 0.866600790513834, "confidence_cutoff": 0.5841912218049162 } }, "score_mae": 0.4606092452508017, "ranked_probability_score": 0.07134033027857928 }, "none_absent": { "n": 60, "nll": 0.543251177325835, "acc": 0.85, "ece": 0.03515603322900176, "brier": 0.22331079271298393, "mean_conf": 0.8600576005410744, "confident_error_rate": 0.03333333333333333, "coverage_at_0_9": 0.6333333333333333, "accuracy_at_0_9": 0.9473684210526315, "coverage_at_5pct_error": 0.7166666666666667, "coverage_at_1pct_error": 0.0, "aurc": 0.11368855700250266, "error_rate_at_0_9": 0.05263157894736842, "confidence_bias": 0.010057600541074407, "top_bins": { "0.9": { "n": 38, "errors": 2, "error_rate": 0.05263157894736842 }, "0.95": { "n": 31, "errors": 2, "error_rate": 0.06451612903225806 }, "0.99": { "n": 18, "errors": 1, "error_rate": 0.05555555555555555 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.9333333333333333, "confidence_cutoff": 0.9619653601083108 }, "0.8": { "coverage": 0.8, "accuracy": 0.9166666666666666, "confidence_cutoff": 0.683343371252518 } } }, "none_present": { "n": 60, "nll": 0.9005504316398635, "acc": 0.7333333333333333, "ece": 0.09230758815850672, "brier": 0.32249124714796495, "mean_conf": 0.8206757401298138, "confident_error_rate": 0.016666666666666666, "coverage_at_0_9": 0.5666666666666667, "accuracy_at_0_9": 0.9705882352941176, "coverage_at_5pct_error": 0.5833333333333334, "coverage_at_1pct_error": 0.016666666666666666, "aurc": 0.11577228698943155, "error_rate_at_0_9": 0.029411764705882353, "confidence_bias": 0.08734240679648053, "top_bins": { "0.9": { "n": 34, "errors": 1, "error_rate": 0.029411764705882353 }, "0.95": { "n": 27, "errors": 1, "error_rate": 0.037037037037037035 }, "0.99": { "n": 15, "errors": 1, "error_rate": 0.06666666666666667 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.9666666666666667, "confidence_cutoff": 0.9339339032600245 }, "0.8": { "coverage": 0.8, "accuracy": 0.8541666666666666, "confidence_cutoff": 0.6044387108750882 } } }, "permuted": { "n": 84, "nll": 0.4678339065572589, "acc": 0.9166666666666666, "ece": 0.03085079906002683, "brier": 0.1315219130880203, "mean_conf": 0.9279063634937833, "confident_error_rate": 0.023809523809523808, "coverage_at_0_9": 0.7976190476190477, "accuracy_at_0_9": 0.9701492537313433, "coverage_at_5pct_error": 0.9047619047619048, "coverage_at_1pct_error": 0.09523809523809523, "aurc": 0.03856923320884028, "error_rate_at_0_9": 0.029850746268656716, "confidence_bias": 0.011239696827116674, "top_bins": { "0.9": { "n": 67, "errors": 2, "error_rate": 0.029850746268656716 }, "0.95": { "n": 60, "errors": 2, "error_rate": 0.03333333333333333 }, "0.99": { "n": 48, "errors": 1, "error_rate": 0.020833333333333332 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.9761904761904762, "confidence_cutoff": 0.995227212706841 }, "0.8": { "coverage": 0.8095238095238095, "accuracy": 0.9705882352941176, "confidence_cutoff": 0.8928713597792232 } } } }, "heldout_tasks": {}, "permutation": { "n": 60, "mean_max_delta": 0.03779500413383731, "flip_rate": 0.016666666666666666 }, "temperature": 1.319507910772894, "calibrated_clean": { "n": 1264, "nll": 0.4919571472502258, "acc": 0.7943037974683544, "ece": 0.028814258275892082, "brier": 0.2749206448948224, "mean_conf": 0.7989016595250444, "confident_error_rate": 0.020569620253164556, "coverage_at_0_9": 0.5055379746835443, "accuracy_at_0_9": 0.9593114241001565, "coverage_at_5pct_error": 0.5545886075949367, "coverage_at_1pct_error": 0.005537974683544304, "aurc": 0.07095897819981735, "error_rate_at_0_9": 0.0406885758998435, "confidence_bias": 0.004597862056689994, "top_bins": { "0.9": { "n": 639, "errors": 26, "error_rate": 0.0406885758998435 }, "0.95": { "n": 558, "errors": 18, "error_rate": 0.03225806451612903 }, "0.99": { "n": 367, "errors": 6, "error_rate": 0.01634877384196185 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.9588607594936709, "confidence_cutoff": 0.9097529357933202 }, "0.8": { "coverage": 0.8006329113924051, "accuracy": 0.8656126482213439, "confidence_cutoff": 0.5412959766037492 } }, "score_mae": 0.47272708593819684, "ranked_probability_score": 0.07114286982306338 }, "metric_policy": { "version": 2, "selective_ties": "whole_confidence_groups", "coverage_at_error": "in-sample maximum over confidence thresholds; not a deployed error guarantee", "aurc": "right-step integral over whole confidence groups", "confident_error_rate": "high-confidence errors divided by all questions", "error_rate_at_0_9": "errors divided by questions accepted at p_max >= 0.9", "nll": "exact from logits when recorded; otherwise from floored probabilities", "nll_floor": 1e-09, "renormalize_returned_probabilities": true, "raw_sums_outside_1e_5": 0, "returned_zeros": 0 }, "coverage": { "requested_records": 1204, "requested_questions": 1468, "evaluated_records": 1204, "evaluated_questions": 1468, "rejected_records": 0, "truncated_records": 0 }, "latency_ms": { "median": 343.6476665083319, "p95": 1776.4106375863773 }, "calibration": { "inference_temperature": 1.0, "additional_temperature": 1.319507910772894, "logits_recorded": true }, "transfer": { "objective": -0.9407704885894344, "paired_flip": { "pairs": 64, "flip_rate": 0.0625, "both_correct_rate": 0.015625, "invariant_pairs": 24, "invariance_rate": 0.9166666666666666, "invariant_both_correct_rate": 0.5 }, "unknowable": null, "clean": { "n": 656, "nll": 0.9098667547927495, "acc": 0.5685975609756098, "ece": 0.15184069475083165, "brier": 0.5496946473177795, "mean_conf": 0.7148065244031183, "confident_error_rate": 0.06402439024390244, "coverage_at_0_9": 0.3003048780487805, "accuracy_at_0_9": 0.7868020304568528, "coverage_at_5pct_error": 0.13414634146341464, "coverage_at_1pct_error": 0.0701219512195122, "aurc": 0.2726527308927074, "error_rate_at_0_9": 0.2131979695431472, "confidence_bias": 0.14620896342750855, "top_bins": { "0.9": { "n": 197, "errors": 42, "error_rate": 0.2131979695431472 }, "0.95": { "n": 126, "errors": 14, "error_rate": 0.1111111111111111 }, "0.99": { "n": 39, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.6707317073170732, "confidence_cutoff": 0.707867643471588 }, "0.8": { "coverage": 0.8003048780487805, "accuracy": 0.6, "confidence_cutoff": 0.5144661954363468 } }, "score_mae": 0.6961418739313883, "ranked_probability_score": 0.30817160000540955 }, "tasks": { "composition_held_and_or": { "n": 32, "nll": 0.6833422496464963, "acc": 0.5625, "ece": 0.048464233979846005, "brier": 0.4902079366809463, "mean_conf": 0.514035766020154, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.09375, "coverage_at_1pct_error": 0.09375, "aurc": 0.29136050867534974, "error_rate_at_0_9": null, "confidence_bias": -0.048464233979846005, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.625, "confidence_cutoff": 0.5107448405471349 }, "0.8": { "coverage": 0.84375, "accuracy": 0.5925925925925926, "confidence_cutoff": 0.503829673046758 } } }, "composition_held_conditional": { "n": 32, "nll": 0.6940730345557493, "acc": 0.46875, "ece": 0.04352010198047873, "brier": 0.5009236559654398, "mean_conf": 0.5122701019804787, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.0, "coverage_at_1pct_error": 0.0, "aurc": 0.5496196217669959, "error_rate_at_0_9": null, "confidence_bias": 0.04352010198047873, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.5, "confidence_cutoff": 0.5166306187730659 }, "0.8": { "coverage": 0.8125, "accuracy": 0.4230769230769231, "confidence_cutoff": 0.501518502876714 } } }, "composition_held_or_not": { "n": 32, "nll": 0.6980906968444973, "acc": 0.46875, "ece": 0.054383457719069384, "brier": 0.5049256050377603, "mean_conf": 0.5231334577190694, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.0, "coverage_at_1pct_error": 0.0, "aurc": 0.5820981501169419, "error_rate_at_0_9": null, "confidence_bias": 0.054383457719069384, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.375, "confidence_cutoff": 0.5229583382606506 }, "0.8": { "coverage": 0.8125, "accuracy": 0.5769230769230769, "confidence_cutoff": 0.5096097144368206 } } }, "contrastive_authorization": { "n": 40, "nll": 1.0339701294493764, "acc": 0.5, "ece": 0.4144774026307121, "brier": 0.7161675205460087, "mean_conf": 0.9144774026307122, "confident_error_rate": 0.2, "coverage_at_0_9": 0.7, "accuracy_at_0_9": 0.7142857142857143, "coverage_at_5pct_error": 0.525, "coverage_at_1pct_error": 0.5, "aurc": 0.15959830910365297, "error_rate_at_0_9": 0.2857142857142857, "confidence_bias": 0.4144774026307122, "top_bins": { "0.9": { "n": 28, "errors": 8, "error_rate": 0.2857142857142857 }, "0.95": { "n": 21, "errors": 1, "error_rate": 0.047619047619047616 }, "0.99": { "n": 12, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.9799628239498235 }, "0.8": { "coverage": 0.8, "accuracy": 0.625, "confidence_cutoff": 0.8601182103157043 } } }, "contrastive_deadline": { "n": 40, "nll": 2.2505586636609918, "acc": 0.275, "ece": 0.6290899198958707, "brier": 1.197235252147983, "mean_conf": 0.9040899198958708, "confident_error_rate": 0.475, "coverage_at_0_9": 0.7, "accuracy_at_0_9": 0.32142857142857145, "coverage_at_5pct_error": 0.05, "coverage_at_1pct_error": 0.05, "aurc": 0.54830887260108, "error_rate_at_0_9": 0.6785714285714286, "confidence_bias": 0.6290899198958708, "top_bins": { "0.9": { "n": 28, "errors": 19, "error_rate": 0.6785714285714286 }, "0.95": { "n": 9, "errors": 4, "error_rate": 0.4444444444444444 }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.45, "confidence_cutoff": 0.9255054173738333 }, "0.8": { "coverage": 0.8, "accuracy": 0.3125, "confidence_cutoff": 0.8918043030731547 } }, "score_mae": 0.6961418739313883, "ranked_probability_score": 0.30817160000540955 }, "emotion": { "n": 80, "nll": 1.2057542293509105, "acc": 0.5625, "ece": 0.06829600936372701, "brier": 0.5855855421120933, "mean_conf": 0.5838921344491043, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0125, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 0.05, "coverage_at_1pct_error": 0.05, "aurc": 0.2847309733715896, "error_rate_at_0_9": 0.0, "confidence_bias": 0.02139213444910426, "top_bins": { "0.9": { "n": 1, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.7, "confidence_cutoff": 0.5569462304170179 }, "0.8": { "coverage": 0.8, "accuracy": 0.625, "confidence_cutoff": 0.4726999258969467 } } }, "mmlu": { "n": 80, "nll": 1.1222459884756266, "acc": 0.525, "ece": 0.12375383831039756, "brier": 0.5704439813752953, "mean_conf": 0.5562091193375563, "confident_error_rate": 0.0, "coverage_at_0_9": 0.1125, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 0.25, "coverage_at_1pct_error": 0.2375, "aurc": 0.2604475173843487, "error_rate_at_0_9": 0.0, "confidence_bias": 0.031209119337556324, "top_bins": { "0.9": { "n": 9, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 5, "errors": 0, "error_rate": 0.0 }, "0.99": { "n": 1, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.7, "confidence_cutoff": 0.5009298640715034 }, "0.8": { "coverage": 0.8, "accuracy": 0.5625, "confidence_cutoff": 0.3604842722415924 } } }, "paws": { "n": 80, "nll": 0.9816472399168571, "acc": 0.2875, "ece": 0.35924160997690124, "brier": 0.7544718149675231, "mean_conf": 0.6467416099769012, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.0, "coverage_at_1pct_error": 0.0, "aurc": 0.8563656691854259, "error_rate_at_0_9": null, "confidence_bias": 0.35924160997690124, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.1, "confidence_cutoff": 0.6299932786820316 }, "0.8": { "coverage": 0.8, "accuracy": 0.25, "confidence_cutoff": 0.5468162135238583 } } }, "qnli": { "n": 80, "nll": 0.6201317262227193, "acc": 0.6875, "ece": 0.11840502530533466, "brier": 0.41161498583927425, "mean_conf": 0.7757451323961314, "confident_error_rate": 0.0375, "coverage_at_0_9": 0.2875, "accuracy_at_0_9": 0.8695652173913043, "coverage_at_5pct_error": 0.025, "coverage_at_1pct_error": 0.025, "aurc": 0.21007195888747154, "error_rate_at_0_9": 0.13043478260869565, "confidence_bias": 0.08824513239613141, "top_bins": { "0.9": { "n": 23, "errors": 3, "error_rate": 0.13043478260869565 }, "0.95": { "n": 10, "errors": 2, "error_rate": 0.2 }, "0.99": { "n": 1, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.775, "confidence_cutoff": 0.8032595395556161 }, "0.8": { "coverage": 0.8, "accuracy": 0.765625, "confidence_cutoff": 0.6199027115135577 } } }, "sciq": { "n": 80, "nll": 0.22354972299184625, "acc": 0.925, "ece": 0.06789633455655048, "brier": 0.12012071486848477, "mean_conf": 0.8789441216334387, "confident_error_rate": 0.0, "coverage_at_0_9": 0.675, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 0.9, "coverage_at_1pct_error": 0.75, "aurc": 0.01049294556907073, "error_rate_at_0_9": 0.0, "confidence_bias": -0.04605587836656133, "top_bins": { "0.9": { "n": 54, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 47, "errors": 0, "error_rate": 0.0 }, "0.99": { "n": 22, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.9726309307076344 }, "0.8": { "coverage": 0.8, "accuracy": 0.984375, "confidence_cutoff": 0.7994060744804564 } } }, "tweet_offensive": { "n": 80, "nll": 0.8351116933687053, "acc": 0.6875, "ece": 0.22180319280009556, "brier": 0.5101348034224664, "mean_conf": 0.8908219907612654, "confident_error_rate": 0.15, "coverage_at_0_9": 0.675, "accuracy_at_0_9": 0.7777777777777778, "coverage_at_5pct_error": 0.0625, "coverage_at_1pct_error": 0.0625, "aurc": 0.20980111038804497, "error_rate_at_0_9": 0.2222222222222222, "confidence_bias": 0.20332199076126545, "top_bins": { "0.9": { "n": 54, "errors": 12, "error_rate": 0.2222222222222222 }, "0.95": { "n": 34, "errors": 7, "error_rate": 0.20588235294117646 }, "0.99": { "n": 3, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.775, "confidence_cutoff": 0.9442175626754761 }, "0.8": { "coverage": 0.8, "accuracy": 0.703125, "confidence_cutoff": 0.8313192624640026 } } } }, "variants": { "clean": { "n": 656, "nll": 0.9098667547927495, "acc": 0.5685975609756098, "ece": 0.15184069475083165, "brier": 0.5496946473177795, "mean_conf": 0.7148065244031183, "confident_error_rate": 0.06402439024390244, "coverage_at_0_9": 0.3003048780487805, "accuracy_at_0_9": 0.7868020304568528, "coverage_at_5pct_error": 0.13414634146341464, "coverage_at_1pct_error": 0.0701219512195122, "aurc": 0.2726527308927074, "error_rate_at_0_9": 0.2131979695431472, "confidence_bias": 0.14620896342750855, "top_bins": { "0.9": { "n": 197, "errors": 42, "error_rate": 0.2131979695431472 }, "0.95": { "n": 126, "errors": 14, "error_rate": 0.1111111111111111 }, "0.99": { "n": 39, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.6707317073170732, "confidence_cutoff": 0.707867643471588 }, "0.8": { "coverage": 0.8003048780487805, "accuracy": 0.6, "confidence_cutoff": 0.5144661954363468 } }, "score_mae": 0.6961418739313883, "ranked_probability_score": 0.30817160000540955 }, "none_absent": { "n": 36, "nll": 0.839222556790093, "acc": 0.6388888888888888, "ece": 0.13519060875689604, "brier": 0.4790985720376279, "mean_conf": 0.6663556845610538, "confident_error_rate": 0.027777777777777776, "coverage_at_0_9": 0.16666666666666666, "accuracy_at_0_9": 0.8333333333333334, "coverage_at_5pct_error": 0.08333333333333333, "coverage_at_1pct_error": 0.08333333333333333, "aurc": 0.24216988587274996, "error_rate_at_0_9": 0.16666666666666666, "confidence_bias": 0.027466795672164923, "top_bins": { "0.9": { "n": 6, "errors": 1, "error_rate": 0.16666666666666666 }, "0.95": { "n": 5, "errors": 1, "error_rate": 0.2 }, "0.99": { "n": 1, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.7222222222222222, "confidence_cutoff": 0.692634932148079 }, "0.8": { "coverage": 0.8055555555555556, "accuracy": 0.7586206896551724, "confidence_cutoff": 0.46444892796790427 } } }, "none_present": { "n": 36, "nll": 1.0702554096724084, "acc": 0.6111111111111112, "ece": 0.0824644297325398, "brier": 0.4845346826960828, "mean_conf": 0.6615826453720488, "confident_error_rate": 0.0, "coverage_at_0_9": 0.2222222222222222, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 0.3888888888888889, "coverage_at_1pct_error": 0.3888888888888889, "aurc": 0.13647701567895199, "error_rate_at_0_9": 0.0, "confidence_bias": 0.050471534260937645, "top_bins": { "0.9": { "n": 8, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 5, "errors": 0, "error_rate": 0.0 }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.9444444444444444, "confidence_cutoff": 0.678112416947823 }, "0.8": { "coverage": 0.8055555555555556, "accuracy": 0.6896551724137931, "confidence_cutoff": 0.4202956126164172 } } }, "permuted": { "n": 36, "nll": 0.7676115841056179, "acc": 0.6944444444444444, "ece": 0.13343139218862612, "brier": 0.3608910203704398, "mean_conf": 0.7179161874094808, "confident_error_rate": 0.0, "coverage_at_0_9": 0.3055555555555556, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 0.5555555555555556, "coverage_at_1pct_error": 0.5277777777777778, "aurc": 0.08668255163693558, "error_rate_at_0_9": 0.0, "confidence_bias": 0.023471742965036402, "top_bins": { "0.9": { "n": 11, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 9, "errors": 0, "error_rate": 0.0 }, "0.99": { "n": 4, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.7402256166503357 }, "0.8": { "coverage": 0.8055555555555556, "accuracy": 0.7931034482758621, "confidence_cutoff": 0.5095580033363343 } } } }, "heldout_tasks": { "composition_held_and_or": { "n": 32, "nll": 0.6833422496464963, "acc": 0.5625, "ece": 0.048464233979846005, "brier": 0.4902079366809463, "mean_conf": 0.514035766020154, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.09375, "coverage_at_1pct_error": 0.09375, "aurc": 0.29136050867534974, "error_rate_at_0_9": null, "confidence_bias": -0.048464233979846005, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.625, "confidence_cutoff": 0.5107448405471349 }, "0.8": { "coverage": 0.84375, "accuracy": 0.5925925925925926, "confidence_cutoff": 0.503829673046758 } } }, "composition_held_conditional": { "n": 32, "nll": 0.6940730345557493, "acc": 0.46875, "ece": 0.04352010198047873, "brier": 0.5009236559654398, "mean_conf": 0.5122701019804787, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.0, "coverage_at_1pct_error": 0.0, "aurc": 0.5496196217669959, "error_rate_at_0_9": null, "confidence_bias": 0.04352010198047873, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.5, "confidence_cutoff": 0.5166306187730659 }, "0.8": { "coverage": 0.8125, "accuracy": 0.4230769230769231, "confidence_cutoff": 0.501518502876714 } } }, "composition_held_or_not": { "n": 32, "nll": 0.6980906968444973, "acc": 0.46875, "ece": 0.054383457719069384, "brier": 0.5049256050377603, "mean_conf": 0.5231334577190694, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.0, "coverage_at_1pct_error": 0.0, "aurc": 0.5820981501169419, "error_rate_at_0_9": null, "confidence_bias": 0.054383457719069384, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.375, "confidence_cutoff": 0.5229583382606506 }, "0.8": { "coverage": 0.8125, "accuracy": 0.5769230769230769, "confidence_cutoff": 0.5096097144368206 } } }, "contrastive_authorization": { "n": 40, "nll": 1.0339701294493764, "acc": 0.5, "ece": 0.4144774026307121, "brier": 0.7161675205460087, "mean_conf": 0.9144774026307122, "confident_error_rate": 0.2, "coverage_at_0_9": 0.7, "accuracy_at_0_9": 0.7142857142857143, "coverage_at_5pct_error": 0.525, "coverage_at_1pct_error": 0.5, "aurc": 0.15959830910365297, "error_rate_at_0_9": 0.2857142857142857, "confidence_bias": 0.4144774026307122, "top_bins": { "0.9": { "n": 28, "errors": 8, "error_rate": 0.2857142857142857 }, "0.95": { "n": 21, "errors": 1, "error_rate": 0.047619047619047616 }, "0.99": { "n": 12, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.9799628239498235 }, "0.8": { "coverage": 0.8, "accuracy": 0.625, "confidence_cutoff": 0.8601182103157043 } } }, "contrastive_deadline": { "n": 40, "nll": 2.2505586636609918, "acc": 0.275, "ece": 0.6290899198958707, "brier": 1.197235252147983, "mean_conf": 0.9040899198958708, "confident_error_rate": 0.475, "coverage_at_0_9": 0.7, "accuracy_at_0_9": 0.32142857142857145, "coverage_at_5pct_error": 0.05, "coverage_at_1pct_error": 0.05, "aurc": 0.54830887260108, "error_rate_at_0_9": 0.6785714285714286, "confidence_bias": 0.6290899198958708, "top_bins": { "0.9": { "n": 28, "errors": 19, "error_rate": 0.6785714285714286 }, "0.95": { "n": 9, "errors": 4, "error_rate": 0.4444444444444444 }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.45, "confidence_cutoff": 0.9255054173738333 }, "0.8": { "coverage": 0.8, "accuracy": 0.3125, "confidence_cutoff": 0.8918043030731547 } }, "score_mae": 0.6961418739313883, "ranked_probability_score": 0.30817160000540955 }, "emotion": { "n": 80, "nll": 1.2057542293509105, "acc": 0.5625, "ece": 0.06829600936372701, "brier": 0.5855855421120933, "mean_conf": 0.5838921344491043, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0125, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 0.05, "coverage_at_1pct_error": 0.05, "aurc": 0.2847309733715896, "error_rate_at_0_9": 0.0, "confidence_bias": 0.02139213444910426, "top_bins": { "0.9": { "n": 1, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.7, "confidence_cutoff": 0.5569462304170179 }, "0.8": { "coverage": 0.8, "accuracy": 0.625, "confidence_cutoff": 0.4726999258969467 } } }, "mmlu": { "n": 80, "nll": 1.1222459884756266, "acc": 0.525, "ece": 0.12375383831039756, "brier": 0.5704439813752953, "mean_conf": 0.5562091193375563, "confident_error_rate": 0.0, "coverage_at_0_9": 0.1125, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 0.25, "coverage_at_1pct_error": 0.2375, "aurc": 0.2604475173843487, "error_rate_at_0_9": 0.0, "confidence_bias": 0.031209119337556324, "top_bins": { "0.9": { "n": 9, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 5, "errors": 0, "error_rate": 0.0 }, "0.99": { "n": 1, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.7, "confidence_cutoff": 0.5009298640715034 }, "0.8": { "coverage": 0.8, "accuracy": 0.5625, "confidence_cutoff": 0.3604842722415924 } } }, "paws": { "n": 80, "nll": 0.9816472399168571, "acc": 0.2875, "ece": 0.35924160997690124, "brier": 0.7544718149675231, "mean_conf": 0.6467416099769012, "confident_error_rate": 0.0, "coverage_at_0_9": 0.0, "accuracy_at_0_9": null, "coverage_at_5pct_error": 0.0, "coverage_at_1pct_error": 0.0, "aurc": 0.8563656691854259, "error_rate_at_0_9": null, "confidence_bias": 0.35924160997690124, "top_bins": { "0.9": { "n": 0, "errors": 0, "error_rate": null }, "0.95": { "n": 0, "errors": 0, "error_rate": null }, "0.99": { "n": 0, "errors": 0, "error_rate": null } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.1, "confidence_cutoff": 0.6299932786820316 }, "0.8": { "coverage": 0.8, "accuracy": 0.25, "confidence_cutoff": 0.5468162135238583 } } }, "qnli": { "n": 80, "nll": 0.6201317262227193, "acc": 0.6875, "ece": 0.11840502530533466, "brier": 0.41161498583927425, "mean_conf": 0.7757451323961314, "confident_error_rate": 0.0375, "coverage_at_0_9": 0.2875, "accuracy_at_0_9": 0.8695652173913043, "coverage_at_5pct_error": 0.025, "coverage_at_1pct_error": 0.025, "aurc": 0.21007195888747154, "error_rate_at_0_9": 0.13043478260869565, "confidence_bias": 0.08824513239613141, "top_bins": { "0.9": { "n": 23, "errors": 3, "error_rate": 0.13043478260869565 }, "0.95": { "n": 10, "errors": 2, "error_rate": 0.2 }, "0.99": { "n": 1, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.775, "confidence_cutoff": 0.8032595395556161 }, "0.8": { "coverage": 0.8, "accuracy": 0.765625, "confidence_cutoff": 0.6199027115135577 } } }, "sciq": { "n": 80, "nll": 0.22354972299184625, "acc": 0.925, "ece": 0.06789633455655048, "brier": 0.12012071486848477, "mean_conf": 0.8789441216334387, "confident_error_rate": 0.0, "coverage_at_0_9": 0.675, "accuracy_at_0_9": 1.0, "coverage_at_5pct_error": 0.9, "coverage_at_1pct_error": 0.75, "aurc": 0.01049294556907073, "error_rate_at_0_9": 0.0, "confidence_bias": -0.04605587836656133, "top_bins": { "0.9": { "n": 54, "errors": 0, "error_rate": 0.0 }, "0.95": { "n": 47, "errors": 0, "error_rate": 0.0 }, "0.99": { "n": 22, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 1.0, "confidence_cutoff": 0.9726309307076344 }, "0.8": { "coverage": 0.8, "accuracy": 0.984375, "confidence_cutoff": 0.7994060744804564 } } }, "tweet_offensive": { "n": 80, "nll": 0.8351116933687053, "acc": 0.6875, "ece": 0.22180319280009556, "brier": 0.5101348034224664, "mean_conf": 0.8908219907612654, "confident_error_rate": 0.15, "coverage_at_0_9": 0.675, "accuracy_at_0_9": 0.7777777777777778, "coverage_at_5pct_error": 0.0625, "coverage_at_1pct_error": 0.0625, "aurc": 0.20980111038804497, "error_rate_at_0_9": 0.2222222222222222, "confidence_bias": 0.20332199076126545, "top_bins": { "0.9": { "n": 54, "errors": 12, "error_rate": 0.2222222222222222 }, "0.95": { "n": 34, "errors": 7, "error_rate": 0.20588235294117646 }, "0.99": { "n": 3, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.775, "confidence_cutoff": 0.9442175626754761 }, "0.8": { "coverage": 0.8, "accuracy": 0.703125, "confidence_cutoff": 0.8313192624640026 } } } }, "permutation": { "n": 36, "mean_max_delta": 0.07768421211680508, "flip_rate": 0.08333333333333333 }, "temperature": 1.319507910772894, "calibrated_clean": { "n": 656, "nll": 0.8448515581702054, "acc": 0.5685975609756098, "ece": 0.11046275792106883, "brier": 0.5254015776080041, "mean_conf": 0.6726607987326776, "confident_error_rate": 0.021341463414634148, "coverage_at_0_9": 0.19207317073170732, "accuracy_at_0_9": 0.8888888888888888, "coverage_at_5pct_error": 0.12804878048780488, "coverage_at_1pct_error": 0.0625, "aurc": 0.2767632520963531, "error_rate_at_0_9": 0.1111111111111111, "confidence_bias": 0.10406323775706783, "top_bins": { "0.9": { "n": 126, "errors": 14, "error_rate": 0.1111111111111111 }, "0.95": { "n": 67, "errors": 3, "error_rate": 0.04477611940298507 }, "0.99": { "n": 12, "errors": 0, "error_rate": 0.0 } }, "selective": { "0.5": { "coverage": 0.5, "accuracy": 0.6646341463414634, "confidence_cutoff": 0.6451680660218214 }, "0.8": { "coverage": 0.8003048780487805, "accuracy": 0.6114285714285714, "confidence_cutoff": 0.5037052906313757 } }, "score_mae": 0.6894445881767614, "ranked_probability_score": 0.28422154802672034 }, "metric_policy": { "version": 2, "selective_ties": "whole_confidence_groups", "coverage_at_error": "in-sample maximum over confidence thresholds; not a deployed error guarantee", "aurc": "right-step integral over whole confidence groups", "confident_error_rate": "high-confidence errors divided by all questions", "error_rate_at_0_9": "errors divided by questions accepted at p_max >= 0.9", "nll": "exact from logits when recorded; otherwise from floored probabilities", "nll_floor": 1e-09, "renormalize_returned_probabilities": true, "raw_sums_outside_1e_5": 0, "returned_zeros": 0 }, "coverage": { "requested_records": 764, "requested_questions": 764, "evaluated_records": 764, "evaluated_questions": 764, "rejected_records": 0, "truncated_records": 0 }, "latency_ms": { "median": 400.28454147977754, "p95": 713.417099986691 }, "calibration": { "inference_temperature": 1.0, "additional_temperature": 1.319507910772894, "logits_recorded": true }, "suite_sha256": "31677c2256b406222e7d94ffdc0a02a70ce05746b9efe307876024c4e77291d1" }, "provenance": { "config": { "base": "google/gemma-4-E2B", "base_revision": "d29ff6b45f081a49ee2733a859c9c9c2d95d1a6f", "seed": 0, "epochs": 1, "lr": 0.0001, "batch": 4, "accum": 2, "dtype": "bf16", "weights_dtype": "bf16", "checkpointing": 1, "p_none_pair": 0.25 }, "config_sha256": "355241763405ebfaa97b051898d14a98695c5ef0ed47b6877082fca47adf1506", "suite_sha256": "a8f50e481b7d90b97da049e0ff6a01cee2f1ed204aed61a8265af0edbb5514d2", "source_hashes": { "kev/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "kev/anchors.py": "6963eafdb276db5a6c94d939eae675c761555553b8d0f963ffd803448beaabbb", "kev/api.py": "7bffacfb762c626b8bc2f670f350295af5ccb0883dbe90239c7d2f8e5ef56582", "kev/autoresearch.py": "f7d7fabb9c0df7065bee3fec4aa4bee7028c5d1565847aefe2d74e3d7d40ed8a", "kev/benchmark.py": "9c442efd8b6b852c574a646192c61ee91afa1306e3b2978bb6391ae1b91a3fa0", "kev/calibrate.py": "c1eff4744fabd349e8abca86777a7aa0cbea44904c8223d3b61fca3d43741519", "kev/checkpoint.py": "9f1dbc5d3d7d3050aca4ee34e0578c9ad76b7a2a7467dfc7a4ad65e3c471db59", "kev/compare.py": "3606dbf98bb305d66420158498cb837304d2edb8dab01cdaf2ec42964bef3435", "kev/composition.py": "f335ed17e18e0a544893db5e22b9a059e6ce1b2e14dbb863ac7d7bbf8f3e0536", "kev/contrastive.py": "cbb979aa5d40265ad0e64695f94b281d91751fa405811ddfa8212fede111edcf", "kev/data.py": "ef1be396e7218ba5c94f0f67de2ceff23a39656b57199cb3aa3192cde53e8320", "kev/device.py": "d1677fd98ec0979c7284546306e34e0d09298ca4042fc39eb2b32a74f4e975c4", "kev/evaluate.py": "999264f2837dcfbf2ec93601aa4745e701698ab850a43ad89324672a6604965c", "kev/experiment.py": "d12ab7f0765de0f29364f3411b99f06bb80b1cbf120ca9ed2b03316ec766e828", "kev/jev.py": "acd4cc3f1844e438cc83a8d409c15ef78a5ef64d39ea10f583d75a6c7646b243", "kev/metrics.py": "dba8d90550999edcd642581fb4da0367bd4b39d08471815184fa732d11e137b9", "kev/mlx_model.py": "f582428796faf6bf962b772a69a6227ac3259ccbd3215c0c8b55ce28dbbd0909", "kev/model.py": "c2129026ab56395f78c7d3071410541d5423dd7d97f7371518570572f206c376", "kev/plot.py": "0874bcff2885d8155a1cceade0de8a275a7163d3c3fd6aa0c294e8ccf5ff7e02", "kev/predictors.py": "b2a66dd9f0f0a8026bc94603383bacfdd622e75d1c935173997175606626a9b3", "kev/publish.py": "8abc9bcf4a01365697b05cf3c5ad0013462bd92d005f5616954e40ab1100c7aa", "kev/serve.py": "c210b1f4598e64b49c399f73603ed372f4314c165ceda50e1340c0466ec1c735", "kev/study_v3.py": "7891150aa623e4479c7c789d4dc4662f186132b37b7bac1e1f072b9e08ee13e3", "kev/suite.py": "d18d5f4f02d079344d61c2a31882c524c90012ebfdb660bf9160e145f37e3d42", "kev/train.py": "e103e99d3a68ab99eb58f135320fbc46b18514fcc2b195d1c6879633836e99b9", "kev/transfer_v9.py": "0902409742151250a28af1fd8f42258b70fcc161f3deed3ee35fe3f473f3c763", "modal_app.py": "7a13e9d29b3fc68f8236a0c1e4d67079b40a2a328a296f7ced1022817dff82aa", "pyproject.toml": "7c17fbe9efcadda3eb488b59adf1b6db5938cd33af4d3bd389ce4359b73fcc7d", "uv.lock": "a9922dbb89acdef78299fd2b4a8c3f7f0fa1b2bc08b55595b6926fa785a9c466" }, "git_commit": "ccd6e8b646068a6689792d76e3cae3698e9a015a", "platform": "HF Jobs l4x1 (job 6ab426ec52d0dbd7f1d86bac)", "torch": "2.8.0+cu128", "device": "cuda", "gpu": "NVIDIA L4", "legacy_checkpoint": false, "measured_checkpoint": { "requested": "runs/g4-e2b-e1-fp32/00-trial-0/checkpoint", "resolved": "runs/g4-e2b-e1-fp32/00-trial-0/checkpoint", "head_sha256": "80bcdddf6f671e9901e3805f8c311492c93d86ac63af0440d519b2d45bb90361", "adapter_sha256": "93a489ba6fd4eddb766e9f9bbd70fbec04babe33c7ecd3925f5f9973e325410a", "inference_temperature": 1.0 }, "eval_device": "mps", "resumed_evaluation": true, "resumed_source_hashes": { "kev/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "kev/anchors.py": "6963eafdb276db5a6c94d939eae675c761555553b8d0f963ffd803448beaabbb", "kev/api.py": "7bffacfb762c626b8bc2f670f350295af5ccb0883dbe90239c7d2f8e5ef56582", "kev/autoresearch.py": "f7d7fabb9c0df7065bee3fec4aa4bee7028c5d1565847aefe2d74e3d7d40ed8a", "kev/benchmark.py": "9c442efd8b6b852c574a646192c61ee91afa1306e3b2978bb6391ae1b91a3fa0", "kev/calibrate.py": "c1eff4744fabd349e8abca86777a7aa0cbea44904c8223d3b61fca3d43741519", "kev/checkpoint.py": "9f1dbc5d3d7d3050aca4ee34e0578c9ad76b7a2a7467dfc7a4ad65e3c471db59", "kev/compare.py": "3606dbf98bb305d66420158498cb837304d2edb8dab01cdaf2ec42964bef3435", "kev/composition.py": "f335ed17e18e0a544893db5e22b9a059e6ce1b2e14dbb863ac7d7bbf8f3e0536", "kev/contrastive.py": "cbb979aa5d40265ad0e64695f94b281d91751fa405811ddfa8212fede111edcf", "kev/data.py": "ef1be396e7218ba5c94f0f67de2ceff23a39656b57199cb3aa3192cde53e8320", "kev/device.py": "d1677fd98ec0979c7284546306e34e0d09298ca4042fc39eb2b32a74f4e975c4", "kev/evaluate.py": "999264f2837dcfbf2ec93601aa4745e701698ab850a43ad89324672a6604965c", "kev/experiment.py": "d12ab7f0765de0f29364f3411b99f06bb80b1cbf120ca9ed2b03316ec766e828", "kev/jev.py": "acd4cc3f1844e438cc83a8d409c15ef78a5ef64d39ea10f583d75a6c7646b243", "kev/metrics.py": "dba8d90550999edcd642581fb4da0367bd4b39d08471815184fa732d11e137b9", "kev/mlx_model.py": "f582428796faf6bf962b772a69a6227ac3259ccbd3215c0c8b55ce28dbbd0909", "kev/model.py": "c2129026ab56395f78c7d3071410541d5423dd7d97f7371518570572f206c376", "kev/plot.py": "0874bcff2885d8155a1cceade0de8a275a7163d3c3fd6aa0c294e8ccf5ff7e02", "kev/predictors.py": "b2a66dd9f0f0a8026bc94603383bacfdd622e75d1c935173997175606626a9b3", "kev/publish.py": "8abc9bcf4a01365697b05cf3c5ad0013462bd92d005f5616954e40ab1100c7aa", "kev/serve.py": "c210b1f4598e64b49c399f73603ed372f4314c165ceda50e1340c0466ec1c735", "kev/study_v3.py": "7891150aa623e4479c7c789d4dc4662f186132b37b7bac1e1f072b9e08ee13e3", "kev/suite.py": "d18d5f4f02d079344d61c2a31882c524c90012ebfdb660bf9160e145f37e3d42", "kev/train.py": "e103e99d3a68ab99eb58f135320fbc46b18514fcc2b195d1c6879633836e99b9", "kev/transfer_v9.py": "0902409742151250a28af1fd8f42258b70fcc161f3deed3ee35fe3f473f3c763", "modal_app.py": "7a13e9d29b3fc68f8236a0c1e4d67079b40a2a328a296f7ced1022817dff82aa", "pyproject.toml": "7c17fbe9efcadda3eb488b59adf1b6db5938cd33af4d3bd389ce4359b73fcc7d", "uv.lock": "a9922dbb89acdef78299fd2b4a8c3f7f0fa1b2bc08b55595b6926fa785a9c466" }, "resumed_with_changed_non_evaluator_files": [], "resumed_git_commit": "ccd6e8b646068a6689792d76e3cae3698e9a015a" }, "mechanism_checks": { "n": 8, "packed_max_delta": 0.0, "sibling_max_delta": 1.8417835235595703e-05, "tolerance": 0.001, "passed": true }, "gates": { "passed": false, "checks": { "complete_coverage": true, "isolation_and_packing": true, "transfer_complete": true, "heldout_pairs_at_least_70pct": false, "transfer_confident_errors_below_10pct": true }, "policy": "Research screening only: 0.001 isolation; 5pp task regression; 70% heldout pair correctness; <=10% confident errors; no automatic release." }, "calibration_fit": { "temperature": 1.319507910772894, "aggregation": "micro", "objective": "raw-logit NLL", "split": "calibration", "rows_sha256": "e8e669ff8b32eca59da46675dbd5a5ec138895480cdd475ed499f8d58249c0a2", "suite_sha256": "a8f50e481b7d90b97da049e0ff6a01cee2f1ed204aed61a8265af0edbb5514d2", "n": 1148 }, "wall_seconds": 1409.0520514999516, "promotable": false, "test_evaluated": false, "training_resources": { "wall_seconds": 4865.690968990326, "records_seen": 15142, "requested_records": 12506, "truncated_records": 0, "rejected_records": 0, "optimizer_steps": 1564, "forward_tokens": 3079220, "peak_device_bytes": 14623992320, "device": "cuda", "dtype": "bf16", "batch": 4, "peak_rss_bytes": 11216687104 } }