nitzotz / eval_results.json
BrainboxAI's picture
Update model
a4fa9f5 verified
Raw History Blame Contribute Delete
16.9 kB
{
"model": "BrainboxAI/nitzotz",
"eval_set": "8 frozen suites, 3990 questions (checksums in the repo's eval/manifest.json)",
"models": {
"nz": "BrainboxAI/nitzotz",
"roeig": "RoeiG/laya-hebrew",
"laya_ml": "convaiinnovations/laya (multilingual)"
},
"licences": {
"nz": "Apache-2.0",
"roeig": "CC-BY-NC-SA-4.0",
"laya_ml": "Apache-2.0"
},
"suites": {
"spam_scam": {
"title": "Scam or not? (298 messages)",
"n": 298,
"chance": 0.5,
"acc": {
"nz": 0.9195,
"roeig": 0.3221,
"laya_ml": 0.5034
},
"brier": {
"nz": 0.1321,
"roeig": 0.7604,
"laya_ml": 0.8042
},
"ece": {
"nz": 0.0714,
"roeig": 0.3802,
"laya_ml": 0.38
},
"empty_acc": {
"nz": 0.7047,
"roeig": 0.2953,
"laya_ml": 0.2953
},
"vs_nitzotz": {
"roeig": {
"nz_only": 186,
"other_only": 8,
"p": 3.5766488810898616e-45
},
"laya_ml": {
"nz_only": 138,
"other_only": 14,
"p": 8.462475337125398e-27
}
}
},
"spam_scam_hard": {
"title": "same, hard cases only (65)",
"n": 65,
"chance": 0.5,
"acc": {
"nz": 0.8308,
"roeig": 0.3231,
"laya_ml": 0.3538
},
"brier": {
"nz": 0.2842,
"roeig": 0.7076,
"laya_ml": 1.1057
},
"ece": {
"nz": 0.1132,
"roeig": 0.3391,
"laya_ml": 0.5564
},
"empty_acc": {
"nz": 0.7077,
"roeig": 0.2923,
"laya_ml": 0.2923
},
"vs_nitzotz": {
"roeig": {
"nz_only": 34,
"other_only": 1,
"p": 2.0954757928848267e-09
},
"laya_ml": {
"nz_only": 35,
"other_only": 4,
"p": 3.353161446284503e-07
}
}
},
"spam_type": {
"title": "Message type, 6 options (298)",
"n": 298,
"chance": 0.1667,
"acc": {
"nz": 0.7785,
"roeig": 0.5268,
"laya_ml": 0.2081
},
"brier": {
"nz": 0.3312,
"roeig": 0.6489,
"laya_ml": 1.0746
},
"ece": {
"nz": 0.0683,
"roeig": 0.1043,
"laya_ml": 0.4131
},
"empty_acc": {
"nz": 0.0772,
"roeig": 0.2013,
"laya_ml": 0.0772
},
"vs_nitzotz": {
"roeig": {
"nz_only": 108,
"other_only": 33,
"p": 1.6862794099851315e-10
},
"laya_ml": {
"nz_only": 177,
"other_only": 7,
"p": 1.0714267651676255e-43
}
}
},
"spam_type_hard": {
"title": "same, hard cases only (65)",
"n": 65,
"chance": 0.1667,
"acc": {
"nz": 0.6769,
"roeig": 0.4462,
"laya_ml": 0.2462
},
"brier": {
"nz": 0.4865,
"roeig": 0.7315,
"laya_ml": 1.1094
},
"ece": {
"nz": 0.1027,
"roeig": 0.1877,
"laya_ml": 0.4109
},
"empty_acc": {
"nz": 0.1385,
"roeig": 0.1538,
"laya_ml": 0.1385
},
"vs_nitzotz": {
"roeig": {
"nz_only": 23,
"other_only": 8,
"p": 0.010673840530216694
},
"laya_ml": {
"nz_only": 31,
"other_only": 3,
"p": 7.660128176212311e-07
}
}
},
"triage_type": {
"title": "Support ticket type, 5 options (30)",
"n": 30,
"chance": 0.2,
"acc": {
"nz": 0.9,
"roeig": 0.8,
"laya_ml": 0.5667
},
"brier": {
"nz": 0.1838,
"roeig": 0.3191,
"laya_ml": 0.6667
},
"ece": {
"nz": 0.1292,
"roeig": 0.1321,
"laya_ml": 0.2772
},
"empty_acc": {
"nz": 0.1667,
"roeig": 0.1667,
"laya_ml": 0.3333
},
"vs_nitzotz": {
"roeig": {
"nz_only": 5,
"other_only": 2,
"p": 0.453125
},
"laya_ml": {
"nz_only": 12,
"other_only": 2,
"p": 0.012939453125
}
}
},
"triage_urgency": {
"title": "Ticket urgency, 5 levels (30)",
"n": 30,
"chance": 0.2,
"acc": {
"nz": 0.7,
"roeig": 0.3667,
"laya_ml": 0.3333
},
"brier": {
"nz": 0.4868,
"roeig": 0.708,
"laya_ml": 0.7501
},
"ece": {
"nz": 0.2041,
"roeig": 0.0322,
"laya_ml": 0.2167
},
"empty_acc": {
"nz": 0.4,
"roeig": 0.2333,
"laya_ml": 0.0667
},
"vs_nitzotz": {
"roeig": {
"nz_only": 14,
"other_only": 4,
"p": 0.0308837890625
},
"laya_ml": {
"nz_only": 15,
"other_only": 4,
"p": 0.0192108154296875
}
}
},
"triage_paying": {
"title": "Paying customer? yes/no (30)",
"n": 30,
"chance": 0.5,
"acc": {
"nz": 0.8,
"roeig": 0.5667,
"laya_ml": 0.5
},
"brier": {
"nz": 0.2841,
"roeig": 0.4338,
"laya_ml": 0.8183
},
"ece": {
"nz": 0.1466,
"roeig": 0.2362,
"laya_ml": 0.4284
},
"empty_acc": {
"nz": 0.5333,
"roeig": 0.5333,
"laya_ml": 0.4667
},
"vs_nitzotz": {
"roeig": {
"nz_only": 7,
"other_only": 0,
"p": 0.015625
},
"laya_ml": {
"nz_only": 14,
"other_only": 5,
"p": 0.063568115234375
}
}
},
"massive_20": {
"title": "Voice command intent, 20 options (500)",
"n": 500,
"chance": 0.05,
"acc": {
"nz": 0.9,
"roeig": 0.726,
"laya_ml": 0.474
},
"brier": {
"nz": 0.1671,
"roeig": 0.3801,
"laya_ml": 0.7338
},
"ece": {
"nz": 0.0583,
"roeig": 0.0395,
"laya_ml": 0.1661
},
"empty_acc": {
"nz": 0.064,
"roeig": 0.052,
"laya_ml": 0.044
},
"vs_nitzotz": {
"roeig": {
"nz_only": 98,
"other_only": 11,
"p": 1.3281749873159828e-18
},
"laya_ml": {
"nz_only": 222,
"other_only": 9,
"p": 2.661469004156119e-54
}
}
},
"massive_4": {
"title": "Voice command intent, 4 options (500)",
"n": 500,
"chance": 0.25,
"acc": {
"nz": 0.972,
"roeig": 0.908,
"laya_ml": 0.694
},
"brier": {
"nz": 0.0462,
"roeig": 0.1369,
"laya_ml": 0.3987
},
"ece": {
"nz": 0.0501,
"roeig": 0.0393,
"laya_ml": 0.0468
},
"empty_acc": {
"nz": 0.254,
"roeig": 0.242,
"laya_ml": 0.256
},
"vs_nitzotz": {
"roeig": {
"nz_only": 35,
"other_only": 3,
"p": 6.677873898297548e-08
},
"laya_ml": {
"nz_only": 141,
"other_only": 2,
"p": 1.846933796755538e-39
}
}
},
"sib200": {
"title": "News topic, 7 options (204)",
"n": 204,
"chance": 0.1429,
"acc": {
"nz": 0.7941,
"roeig": 0.8235,
"laya_ml": 0.6618
},
"brier": {
"nz": 0.3027,
"roeig": 0.285,
"laya_ml": 0.4918
},
"ece": {
"nz": 0.0708,
"roeig": 0.0751,
"laya_ml": 0.1784
},
"empty_acc": {
"nz": 0.1471,
"roeig": 0.1961,
"laya_ml": 0.25
},
"vs_nitzotz": {
"roeig": {
"nz_only": 12,
"other_only": 18,
"p": 0.361594608053565
},
"laya_ml": {
"nz_only": 47,
"other_only": 20,
"p": 0.0013071686857636005
}
}
},
"heq_verify": {
"title": "Does the passage support this answer? (600)",
"n": 600,
"chance": 0.5,
"acc": {
"nz": 0.9483,
"roeig": 0.9417,
"laya_ml": 0.4883
},
"brier": {
"nz": 0.0938,
"roeig": 0.0926,
"laya_ml": 0.8274
},
"ece": {
"nz": 0.0365,
"roeig": 0.0178,
"laya_ml": 0.3946
},
"empty_acc": {
"nz": 0.51,
"roeig": 0.8,
"laya_ml": 0.59
},
"vs_nitzotz": {
"roeig": {
"nz_only": 30,
"other_only": 26,
"p": 0.6888797607233396
},
"laya_ml": {
"nz_only": 296,
"other_only": 20,
"p": 3.520918733970021e-64
}
}
},
"heq_unans": {
"title": "Plausible answer the passage does not give (600)",
"n": 600,
"chance": 0.5,
"acc": {
"nz": 0.8917,
"roeig": 0.53,
"laya_ml": 0.5367
},
"brier": {
"nz": 0.1739,
"roeig": 0.856,
"laya_ml": 0.7348
},
"ece": {
"nz": 0.0484,
"roeig": 0.4336,
"laya_ml": 0.3421
},
"empty_acc": {
"nz": 0.5033,
"roeig": 0.5033,
"laya_ml": 0.5167
},
"vs_nitzotz": {
"roeig": {
"nz_only": 243,
"other_only": 26,
"p": 2.502982180208191e-45
},
"laya_ml": {
"nz_only": 240,
"other_only": 27,
"p": 7.334224519429334e-44
}
}
},
"belebele": {
"title": "Reading comprehension, 4 options (900)",
"n": 900,
"chance": 0.25,
"acc": {
"nz": 0.6322,
"roeig": 0.7544,
"laya_ml": 0.3144
},
"brier": {
"nz": 0.5672,
"roeig": 0.3442,
"laya_ml": 0.8341
},
"ece": {
"nz": 0.1919,
"roeig": 0.0324,
"laya_ml": 0.2307
},
"empty_acc": {
"nz": 0.3011,
"roeig": 0.3189,
"laya_ml": 0.2844
},
"vs_nitzotz": {
"roeig": {
"nz_only": 72,
"other_only": 182,
"p": 3.66511528586527e-12
},
"laya_ml": {
"nz_only": 366,
"other_only": 80,
"p": 9.158467645861244e-45
}
}
}
},
"significance": "paired exact McNemar against Nitzotz on the same questions",
"latency_ms": {
"q8": {
"median": 47.7,
"p90": 75.6,
"n": 50
},
"f16": {
"median": 50.5,
"p90": 77.0,
"n": 50
},
"py": {
"median": 60.0,
"p90": 98.2,
"n": 50
},
"q8_cpu": {
"median": 309.1,
"p90": 711.0,
"n": 50
},
"f16_cpu": {
"median": 425.9,
"p90": 854.0,
"n": 50
},
"py_cpu": {
"median": 223.9,
"p90": 452.7,
"n": 50
}
},
"latency_points_ms": {
"q8": {
"short": {
"item": "massive_he_4-0075",
"input_tokens": 38,
"median_ms": 36.9,
"p90_ms": 38.4,
"min_ms": 35.6
},
"long": {
"item": "heq_unanswerable-0522",
"input_tokens": 430,
"median_ms": 127.4,
"p90_ms": 130.0,
"min_ms": 108.0
}
},
"f16": {
"short": {
"item": "massive_he_4-0075",
"input_tokens": 38,
"median_ms": 43.2,
"p90_ms": 44.3,
"min_ms": 41.2
},
"long": {
"item": "heq_unanswerable-0522",
"input_tokens": 430,
"median_ms": 111.2,
"p90_ms": 118.2,
"min_ms": 108.7
}
},
"py": {
"short": {
"item": "massive_he_4-0075",
"input_tokens": 38,
"median_ms": 43.6,
"p90_ms": 44.7,
"min_ms": 42.4
},
"long": {
"item": "heq_unanswerable-0522",
"input_tokens": 430,
"median_ms": 139.3,
"p90_ms": 157.0,
"min_ms": 137.1
}
},
"q8_cpu": {
"short": {
"item": "massive_he_4-0075",
"input_tokens": 38,
"median_ms": 159.5,
"p90_ms": 185.0,
"min_ms": 117.5
},
"long": {
"item": "heq_unanswerable-0522",
"input_tokens": 430,
"median_ms": 1137.4,
"p90_ms": 1243.6,
"min_ms": 1072.4
}
},
"py_cpu": {
"short": {
"item": "massive_he_4-0075",
"input_tokens": 38,
"median_ms": 112.0,
"p90_ms": 131.3,
"min_ms": 97.1
},
"long": {
"item": "heq_unanswerable-0522",
"input_tokens": 430,
"median_ms": 1021.8,
"p90_ms": 1100.7,
"min_ms": 918.8
}
}
},
"latency_machine": "Intel Core Ultra 9 285H laptop, built-in Arc 140T GPU, Windows 11",
"gguf_parity_50": {
"q8": {
"max_abs_diff": 0.0375,
"mean_abs_diff": 0.001814,
"argmax_agree": "50/50"
},
"f16": {
"max_abs_diff": 0.0017,
"mean_abs_diff": 0.000288,
"argmax_agree": "50/50"
},
"q8_cpu": {
"max_abs_diff": 0.0549,
"mean_abs_diff": 0.003498,
"argmax_agree": "50/50"
},
"f16_cpu": {
"max_abs_diff": 0.0032,
"mean_abs_diff": 0.000278,
"argmax_agree": "50/50"
}
},
"gguf_parity_spam_596": {
"q8": {
"max_abs_diff": 0.0166,
"mean_abs_diff": 0.001297,
"argmax_agree": "596/596",
"accuracy_vs_gold": 0.8473
},
"f16": {
"max_abs_diff": 0.0015,
"mean_abs_diff": 0.000203,
"argmax_agree": "596/596",
"accuracy_vs_gold": 0.8473
},
"py": {
"max_abs_diff": 0.0,
"mean_abs_diff": 0.0,
"argmax_agree": "596/596",
"accuracy_vs_gold": 0.8473
},
"q8_cpu": {
"max_abs_diff": 0.0207,
"mean_abs_diff": 0.001918,
"argmax_agree": "595/596",
"accuracy_vs_gold": 0.8456
},
"f16_cpu": {
"max_abs_diff": 0.0012,
"mean_abs_diff": 0.000185,
"argmax_agree": "596/596",
"accuracy_vs_gold": 0.8473
},
"py_cpu": {
"max_abs_diff": 0.0,
"mean_abs_diff": 0.0,
"argmax_agree": "596/596",
"accuracy_vs_gold": 0.8473
}
},
"belebele_split": {
"verbatim_n": 252,
"reworded_n": 648,
"nz": {
"verbatim": 0.7658730158730159,
"reworded": 0.5802469135802469
},
"roeig": {
"verbatim": 0.8412698412698413,
"reworded": 0.720679012345679
},
"laya_ml": {
"verbatim": 0.44047619047619047,
"reworded": 0.2654320987654321
}
},
"calibration_pooled": {
"n": 3990,
"ece_15bin": 0.0309
},
"wording_robustness": {
"suite": "wording_he (frozen): 1786 questions, each asked in 6 wordings with the same meaning",
"flip_rate": "share of questions whose top answer is not the same in all 6 wordings (lower is better)",
"models": {
"BrainboxAI/nitzotz": {
"all_item_weighted": {
"flip_rate": 0.0504,
"flip_vs_orig": 0.018,
"mean_acc": 0.9118,
"worst_acc": 0.9048,
"flip_rate_spam_type_without_form_swap": 0.0481
},
"parts": {
"heq_verify": {
"n_items": 600,
"mean_acc": 0.9478,
"worst_acc": 0.945,
"flip_rate": 0.0283
},
"massive_he_4": {
"n_items": 500,
"mean_acc": 0.97,
"worst_acc": 0.962,
"flip_rate": 0.036
},
"spam_business_he:category": {
"n_items": 298,
"mean_acc": 0.7752,
"worst_acc": 0.7685,
"flip_rate": 0.0805
},
"spam_business_he:category:hard": {
"n_items": 65,
"mean_acc": 0.6692,
"worst_acc": 0.6462,
"flip_rate": 0.1077
},
"spam_business_he:scam": {
"n_items": 298,
"mean_acc": 0.915,
"worst_acc": 0.9128,
"flip_rate": 0.0201
},
"spam_business_he:scam:hard": {
"n_items": 65,
"mean_acc": 0.8154,
"worst_acc": 0.8,
"flip_rate": 0.0615
},
"triage30:category": {
"n_items": 30,
"mean_acc": 0.9166,
"worst_acc": 0.9,
"flip_rate": 0.0333
},
"triage30:paying": {
"n_items": 30,
"mean_acc": 0.7944,
"worst_acc": 0.7333,
"flip_rate": 0.1667
},
"triage30:urgency": {
"n_items": 30,
"mean_acc": 0.6611,
"worst_acc": 0.6,
"flip_rate": 0.6333
}
},
"single_answer_wordings": {}
}
}
},
"belebele_negation": {
"n": 158,
"acc": 0.5886
},
"seeds": {
"heldout_this": 0.9505,
"others": [
{
"name": "seed 2",
"heldout": 0.9519,
"acc": {
"spam_scam": 0.9195,
"spam_scam_hard": 0.8308,
"spam_type": 0.802,
"spam_type_hard": 0.7077,
"triage_type": 0.9,
"triage_urgency": 0.7333,
"triage_paying": 0.8,
"massive_20": 0.894,
"massive_4": 0.96,
"sib200": 0.8284,
"heq_verify": 0.9467,
"heq_unans": 0.8983,
"belebele": 0.6089
}
},
{
"name": "seed 3",
"heldout": 0.9521,
"acc": {
"spam_scam": 0.906,
"spam_scam_hard": 0.7846,
"spam_type": 0.7785,
"spam_type_hard": 0.6615,
"triage_type": 0.8333,
"triage_urgency": 0.7667,
"triage_paying": 0.8333,
"massive_20": 0.902,
"massive_4": 0.968,
"sib200": 0.8039,
"heq_verify": 0.95,
"heq_unans": 0.8983,
"belebele": 0.6311
}
}
],
"agreement_min": 0.9008,
"agreement_max": 0.9063,
"picked_on": {
"this": {
"n": 188,
"acc": 0.984
},
"others": [
{
"n": 188,
"acc": 0.9734
},
{
"n": 188,
"acc": 0.9734
}
]
}
},
"none_of_the_options": {
"suite": "notinlist_he (frozen): 400 MASSIVE and Belebele test questions with a 'none of the options' choice added; in half of them 'none' is right, in the other half the right answer is still in the list",
"model": "BrainboxAI/nitzotz",
"parts": {
"all": {
"n": 400,
"acc": 0.685,
"picks_none": 0.535
},
"belebele_he/control": {
"n": 100,
"acc": 0.45,
"picks_none": 0.43
},
"belebele_he/none": {
"n": 100,
"acc": 0.76,
"picks_none": 0.76
},
"control": {
"n": 200,
"acc": 0.61,
"picks_none": 0.31
},
"massive_he_20/control": {
"n": 100,
"acc": 0.77,
"picks_none": 0.19
},
"massive_he_20/none": {
"n": 100,
"acc": 0.76,
"picks_none": 0.76
},
"none": {
"n": 200,
"acc": 0.76,
"picks_none": 0.76
}
},
"text_removed": {
"all": {
"n": 400,
"acc": 0.49,
"picks_none": 0.9825
},
"belebele_he/control": {
"n": 100,
"acc": 0.01,
"picks_none": 0.99
},
"belebele_he/none": {
"n": 100,
"acc": 0.95,
"picks_none": 0.95
},
"control": {
"n": 200,
"acc": 0.005,
"picks_none": 0.99
},
"massive_he_20/control": {
"n": 100,
"acc": 0.0,
"picks_none": 0.99
},
"massive_he_20/none": {
"n": 100,
"acc": 1.0,
"picks_none": 1.0
},
"none": {
"n": 200,
"acc": 0.975,
"picks_none": 0.975
}
}
}
}