Files
decision-models-benchmark/reports/test/analysis.json
2026-10-07 20:53:56 -03:00

1944 lines
46 KiB
JSON

{
"split": "test",
"criteria": {
"latency_gpu_batch1": {
"p95_ms_per_decision": 150
},
"accuracy": {
"tasks": [
"D1",
"D2"
],
"non_inferiority_margin_pp": 3.0,
"ci": 0.95,
"bootstrap_unit": "group_id",
"fallback_macro_f1_min": 0.8
},
"risk": {
"max_confident_wrong_route_rate": 0.02,
"min_coverage": 0.8
}
},
"candidates": {
"gliner": {
"cpu": {
"device": "cpu",
"precision": "float32",
"repetitions": 3,
"quality": {
"n": 1744,
"errors": 0,
"accuracy": 0.57,
"accuracy_ci95": [
0.5263,
0.6145
],
"macro_f1": 0.5795,
"macro_precision": 0.7287,
"macro_recall": 0.5628,
"fallback_recall": 0.5873,
"fallback_precision": 0.4029
},
"calibration": {
"temperature": 0.8187,
"raw": {
"brier": 0.5973,
"nll": 1.3014,
"ece": 0.0986
},
"calibrated": {
"brier": 0.5893,
"nll": 1.2793,
"ece": 0.0785
}
},
"abstention": {
"threshold": 0.856167,
"coverage": 0.1279,
"selective_accuracy": 0.8475,
"confident_wrong_rate": 0.0195
},
"latency_ms": {
"p50": {
"mean": 267.31,
"sd": 4.88
},
"p95": {
"mean": 481.32,
"sd": 11.69
},
"p99": {
"mean": 537.18,
"sd": 15.38
},
"mean": {
"mean": 281.08,
"sd": 4.76
}
},
"throughput_decisions_per_s": 3.56,
"cost_usd_per_1k_decisions": 0.0231,
"load_seconds": 19.863333333333333,
"rss_peak_mib": 3275.6171875,
"gpu_mem_peak_mib": null,
"storage_mib": {
"weights": 1112,
"environment": 4115
},
"prediction_agreement_across_reps": 1.0,
"top_confusions": [
{
"gold": "sem_correspondencia",
"pred": "internet_pacote_adicional",
"n": 49
},
{
"gold": "conta_explicacao",
"pred": "sem_correspondencia",
"n": 30
},
{
"gold": "roaming_pacote",
"pred": "internet_pacote_adicional",
"n": 22
},
{
"gold": "marketplace_pedido",
"pred": "sem_correspondencia",
"n": 21
},
{
"gold": "sem_correspondencia",
"pred": "suporte_sem_sinal",
"n": 20
},
{
"gold": "conta_contestacao",
"pred": "sem_correspondencia",
"n": 19
},
{
"gold": "negociacao_divida",
"pred": "sem_correspondencia",
"n": 17
},
{
"gold": "conta_vencimento",
"pred": "sem_correspondencia",
"n": 16
},
{
"gold": "conta_pagamento_confirmacao",
"pred": "sem_correspondencia",
"n": 16
},
{
"gold": "recarga_saldo",
"pred": "sem_correspondencia",
"n": 16
}
],
"error_classes": {
"abstencao_indevida": 329,
"taxonomia_par_dificil": 100,
"intencao_fora_das_opcoes": 88,
"fora_de_escopo_aceito": 66,
"desempenho_do_modelo": 65,
"linguagem": 56,
"ambiguidade": 46
},
"slices": {
"profile": {
"curto_neutro": {
"n": 251,
"accuracy": 0.5618
},
"emocional": {
"n": 309,
"accuracy": 0.6084
},
"erros_digitacao": {
"n": 228,
"accuracy": 0.4956
},
"indireto_multiplo": {
"n": 239,
"accuracy": 0.4812
},
"informal_abreviado": {
"n": 216,
"accuracy": 0.6065
},
"longo_contexto": {
"n": 251,
"accuracy": 0.6016
},
"regionalismo": {
"n": 250,
"accuracy": 0.62
}
},
"n_options": {
"12": {
"n": 422,
"accuracy": 0.5521
},
"16": {
"n": 440,
"accuracy": 0.5136
},
"4": {
"n": 451,
"accuracy": 0.6031
},
"8": {
"n": 431,
"accuracy": 0.6102
}
},
"kind": {
"intent": {
"n": 1508,
"accuracy": 0.5477
},
"oos": {
"n": 236,
"accuracy": 0.7119
}
},
"gold_eligible": {
"False": {
"n": 142,
"accuracy": 0.3803
},
"True": {
"n": 1602,
"accuracy": 0.5868
}
},
"length_bucket": {
"curto": {
"n": 578,
"accuracy": 0.5779
},
"longo": {
"n": 109,
"accuracy": 0.5963
},
"medio": {
"n": 1057,
"accuracy": 0.5629
}
},
"sentiment": {
"ansioso": {
"n": 330,
"accuracy": 0.5727
},
"confuso": {
"n": 3,
"accuracy": 0.6667
},
"frustrado": {
"n": 340,
"accuracy": 0.5735
},
"indignado": {
"n": 122,
"accuracy": 0.5492
},
"irritado": {
"n": 243,
"accuracy": 0.5267
},
"neutro": {
"n": 640,
"accuracy": 0.5844
},
"positivo": {
"n": 66,
"accuracy": 0.5909
}
},
"domain": {
"atendimento": {
"n": 72,
"accuracy": 0.4306
},
"beneficios_fidelidade": {
"n": 179,
"accuracy": 0.7207
},
"contas": {
"n": 219,
"accuracy": 0.4612
},
"fora_de_escopo": {
"n": 236,
"accuracy": 0.7119
},
"internet": {
"n": 108,
"accuracy": 0.7222
},
"linha_cadastro": {
"n": 145,
"accuracy": 0.5586
},
"marketplace": {
"n": 107,
"accuracy": 0.3738
},
"negociacao": {
"n": 72,
"accuracy": 0.2778
},
"plano": {
"n": 108,
"accuracy": 0.6019
},
"recarga": {
"n": 108,
"accuracy": 0.4537
},
"roaming": {
"n": 108,
"accuracy": 0.3148
},
"seguranca": {
"n": 71,
"accuracy": 0.8169
},
"servicos_adicionais": {
"n": 104,
"accuracy": 0.5385
},
"suporte": {
"n": 107,
"accuracy": 0.785
}
},
"annotation": {
"consenso": {
"n": 1723,
"accuracy": 0.5728
},
"desempate": {
"n": 21,
"accuracy": 0.3333
}
}
}
},
"cuda": {
"device": "cuda",
"precision": "float32",
"repetitions": 3,
"quality": {
"n": 1744,
"errors": 0,
"accuracy": 0.57,
"accuracy_ci95": [
0.5263,
0.6145
],
"macro_f1": 0.5795,
"macro_precision": 0.7287,
"macro_recall": 0.5628,
"fallback_recall": 0.5873,
"fallback_precision": 0.4029
},
"calibration": {
"temperature": 0.8187,
"raw": {
"brier": 0.5973,
"nll": 1.3014,
"ece": 0.0986
},
"calibrated": {
"brier": 0.5893,
"nll": 1.2793,
"ece": 0.0785
}
},
"abstention": {
"threshold": 0.857453,
"coverage": 0.1279,
"selective_accuracy": 0.8475,
"confident_wrong_rate": 0.0195
},
"latency_ms": {
"p50": {
"mean": 20.8,
"sd": 0.36
},
"p95": {
"mean": 31.71,
"sd": 0.06
},
"p99": {
"mean": 33.83,
"sd": 0.13
},
"mean": {
"mean": 21.76,
"sd": 0.17
}
},
"throughput_decisions_per_s": 45.96,
"cost_usd_per_1k_decisions": 0.0121,
"load_seconds": 21.646666666666665,
"rss_peak_mib": 3277.203125,
"gpu_mem_peak_mib": 1807.0,
"storage_mib": {
"weights": 1112,
"environment": 4115
},
"prediction_agreement_across_reps": 1.0,
"top_confusions": [
{
"gold": "sem_correspondencia",
"pred": "internet_pacote_adicional",
"n": 49
},
{
"gold": "conta_explicacao",
"pred": "sem_correspondencia",
"n": 30
},
{
"gold": "roaming_pacote",
"pred": "internet_pacote_adicional",
"n": 22
},
{
"gold": "marketplace_pedido",
"pred": "sem_correspondencia",
"n": 21
},
{
"gold": "sem_correspondencia",
"pred": "suporte_sem_sinal",
"n": 20
},
{
"gold": "conta_contestacao",
"pred": "sem_correspondencia",
"n": 19
},
{
"gold": "negociacao_divida",
"pred": "sem_correspondencia",
"n": 17
},
{
"gold": "conta_vencimento",
"pred": "sem_correspondencia",
"n": 16
},
{
"gold": "conta_pagamento_confirmacao",
"pred": "sem_correspondencia",
"n": 16
},
{
"gold": "recarga_saldo",
"pred": "sem_correspondencia",
"n": 16
}
],
"error_classes": {
"abstencao_indevida": 329,
"taxonomia_par_dificil": 100,
"intencao_fora_das_opcoes": 88,
"fora_de_escopo_aceito": 66,
"desempenho_do_modelo": 65,
"linguagem": 56,
"ambiguidade": 46
},
"slices": {
"profile": {
"curto_neutro": {
"n": 251,
"accuracy": 0.5618
},
"emocional": {
"n": 309,
"accuracy": 0.6084
},
"erros_digitacao": {
"n": 228,
"accuracy": 0.4956
},
"indireto_multiplo": {
"n": 239,
"accuracy": 0.4812
},
"informal_abreviado": {
"n": 216,
"accuracy": 0.6065
},
"longo_contexto": {
"n": 251,
"accuracy": 0.6016
},
"regionalismo": {
"n": 250,
"accuracy": 0.62
}
},
"n_options": {
"12": {
"n": 422,
"accuracy": 0.5521
},
"16": {
"n": 440,
"accuracy": 0.5136
},
"4": {
"n": 451,
"accuracy": 0.6031
},
"8": {
"n": 431,
"accuracy": 0.6102
}
},
"kind": {
"intent": {
"n": 1508,
"accuracy": 0.5477
},
"oos": {
"n": 236,
"accuracy": 0.7119
}
},
"gold_eligible": {
"False": {
"n": 142,
"accuracy": 0.3803
},
"True": {
"n": 1602,
"accuracy": 0.5868
}
},
"length_bucket": {
"curto": {
"n": 578,
"accuracy": 0.5779
},
"longo": {
"n": 109,
"accuracy": 0.5963
},
"medio": {
"n": 1057,
"accuracy": 0.5629
}
},
"sentiment": {
"ansioso": {
"n": 330,
"accuracy": 0.5727
},
"confuso": {
"n": 3,
"accuracy": 0.6667
},
"frustrado": {
"n": 340,
"accuracy": 0.5735
},
"indignado": {
"n": 122,
"accuracy": 0.5492
},
"irritado": {
"n": 243,
"accuracy": 0.5267
},
"neutro": {
"n": 640,
"accuracy": 0.5844
},
"positivo": {
"n": 66,
"accuracy": 0.5909
}
},
"domain": {
"atendimento": {
"n": 72,
"accuracy": 0.4306
},
"beneficios_fidelidade": {
"n": 179,
"accuracy": 0.7207
},
"contas": {
"n": 219,
"accuracy": 0.4612
},
"fora_de_escopo": {
"n": 236,
"accuracy": 0.7119
},
"internet": {
"n": 108,
"accuracy": 0.7222
},
"linha_cadastro": {
"n": 145,
"accuracy": 0.5586
},
"marketplace": {
"n": 107,
"accuracy": 0.3738
},
"negociacao": {
"n": 72,
"accuracy": 0.2778
},
"plano": {
"n": 108,
"accuracy": 0.6019
},
"recarga": {
"n": 108,
"accuracy": 0.4537
},
"roaming": {
"n": 108,
"accuracy": 0.3148
},
"seguranca": {
"n": 71,
"accuracy": 0.8169
},
"servicos_adicionais": {
"n": 104,
"accuracy": 0.5385
},
"suporte": {
"n": 107,
"accuracy": 0.785
}
},
"annotation": {
"consenso": {
"n": 1723,
"accuracy": 0.5728
},
"desempate": {
"n": 21,
"accuracy": 0.3333
}
}
}
}
},
"laya": {
"cpu": {
"device": "cpu",
"precision": "float32",
"repetitions": 3,
"quality": {
"n": 1744,
"errors": 0,
"accuracy": 0.5682,
"accuracy_ci95": [
0.5301,
0.6058
],
"macro_f1": 0.5902,
"macro_precision": 0.5858,
"macro_recall": 0.6345,
"fallback_recall": 0.2884,
"fallback_precision": 0.645
},
"calibration": {
"temperature": 3.5787,
"raw": {
"brier": 0.6841,
"nll": 2.2273,
"ece": 0.2689
},
"calibrated": {
"brier": 0.5969,
"nll": 1.3967,
"ece": 0.079
}
},
"abstention": {
"threshold": 0.871532,
"coverage": 0.1044,
"selective_accuracy": 0.8681,
"confident_wrong_rate": 0.0138
},
"latency_ms": {
"p50": {
"mean": 163.61,
"sd": 10.11
},
"p95": {
"mean": 200.86,
"sd": 11.59
},
"p99": {
"mean": 256.99,
"sd": 14.62
},
"mean": {
"mean": 153.95,
"sd": 9.89
}
},
"throughput_decisions_per_s": 6.5,
"cost_usd_per_1k_decisions": 0.0127,
"load_seconds": 8.993333333333334,
"rss_peak_mib": 2689.3359375,
"gpu_mem_peak_mib": null,
"storage_mib": {
"weights": 647,
"environment": 4109
},
"prediction_agreement_across_reps": 1.0,
"top_confusions": [
{
"gold": "sem_correspondencia",
"pred": "cancelamento_linha",
"n": 26
},
{
"gold": "suporte_sem_sinal",
"pred": "internet_problema_conexao",
"n": 25
},
{
"gold": "sem_correspondencia",
"pred": "internet_problema_conexao",
"n": 23
},
{
"gold": "sem_correspondencia",
"pred": "vas_cancelamento",
"n": 14
},
{
"gold": "sem_correspondencia",
"pred": "reclamacao_atendimento",
"n": 14
},
{
"gold": "sem_correspondencia",
"pred": "internet_consumo",
"n": 13
},
{
"gold": "sem_correspondencia",
"pred": "seguranca_fraude",
"n": 13
},
{
"gold": "sem_correspondencia",
"pred": "atendimento_humano",
"n": 13
},
{
"gold": "seguranca_perda_roubo",
"pred": "seguranca_fraude",
"n": 10
},
{
"gold": "portabilidade",
"pred": "troca_numero",
"n": 10
}
],
"error_classes": {
"taxonomia_par_dificil": 182,
"fora_de_escopo_aceito": 139,
"intencao_fora_das_opcoes": 128,
"desempenho_do_modelo": 91,
"linguagem": 83,
"ambiguidade": 70,
"abstencao_indevida": 60
},
"slices": {
"profile": {
"curto_neutro": {
"n": 251,
"accuracy": 0.6056
},
"emocional": {
"n": 309,
"accuracy": 0.5858
},
"erros_digitacao": {
"n": 228,
"accuracy": 0.5658
},
"indireto_multiplo": {
"n": 239,
"accuracy": 0.5314
},
"informal_abreviado": {
"n": 216,
"accuracy": 0.5972
},
"longo_contexto": {
"n": 251,
"accuracy": 0.5578
},
"regionalismo": {
"n": 250,
"accuracy": 0.532
}
},
"n_options": {
"12": {
"n": 422,
"accuracy": 0.5118
},
"16": {
"n": 440,
"accuracy": 0.4795
},
"4": {
"n": 451,
"accuracy": 0.6918
},
"8": {
"n": 431,
"accuracy": 0.5847
}
},
"kind": {
"intent": {
"n": 1508,
"accuracy": 0.5942
},
"oos": {
"n": 236,
"accuracy": 0.4025
}
},
"gold_eligible": {
"False": {
"n": 142,
"accuracy": 0.0915
},
"True": {
"n": 1602,
"accuracy": 0.6105
}
},
"length_bucket": {
"curto": {
"n": 578,
"accuracy": 0.59
},
"longo": {
"n": 109,
"accuracy": 0.5413
},
"medio": {
"n": 1057,
"accuracy": 0.5591
}
},
"sentiment": {
"ansioso": {
"n": 330,
"accuracy": 0.5485
},
"confuso": {
"n": 3,
"accuracy": 0.6667
},
"frustrado": {
"n": 340,
"accuracy": 0.5441
},
"indignado": {
"n": 122,
"accuracy": 0.5574
},
"irritado": {
"n": 243,
"accuracy": 0.5226
},
"neutro": {
"n": 640,
"accuracy": 0.6062
},
"positivo": {
"n": 66,
"accuracy": 0.6061
}
},
"domain": {
"atendimento": {
"n": 72,
"accuracy": 0.5556
},
"beneficios_fidelidade": {
"n": 179,
"accuracy": 0.6592
},
"contas": {
"n": 219,
"accuracy": 0.5479
},
"fora_de_escopo": {
"n": 236,
"accuracy": 0.4068
},
"internet": {
"n": 108,
"accuracy": 0.7222
},
"linha_cadastro": {
"n": 145,
"accuracy": 0.6966
},
"marketplace": {
"n": 107,
"accuracy": 0.5607
},
"negociacao": {
"n": 72,
"accuracy": 0.3611
},
"plano": {
"n": 108,
"accuracy": 0.6019
},
"recarga": {
"n": 108,
"accuracy": 0.7037
},
"roaming": {
"n": 108,
"accuracy": 0.6574
},
"seguranca": {
"n": 71,
"accuracy": 0.3099
},
"servicos_adicionais": {
"n": 104,
"accuracy": 0.5288
},
"suporte": {
"n": 107,
"accuracy": 0.5888
}
},
"annotation": {
"consenso": {
"n": 1723,
"accuracy": 0.567
},
"desempate": {
"n": 21,
"accuracy": 0.6667
}
}
}
},
"cuda": {
"device": "cuda",
"precision": "float32",
"repetitions": 3,
"quality": {
"n": 1744,
"errors": 0,
"accuracy": 0.5677,
"accuracy_ci95": [
0.529,
0.6049
],
"macro_f1": 0.5902,
"macro_precision": 0.5866,
"macro_recall": 0.6343,
"fallback_recall": 0.2857,
"fallback_precision": 0.6391
},
"calibration": {
"temperature": 3.5787,
"raw": {
"brier": 0.6835,
"nll": 2.2246,
"ece": 0.2711
},
"calibrated": {
"brier": 0.5976,
"nll": 1.397,
"ece": 0.0824
}
},
"abstention": {
"threshold": 0.871532,
"coverage": 0.1026,
"selective_accuracy": 0.8659,
"confident_wrong_rate": 0.0138
},
"latency_ms": {
"p50": {
"mean": 21.7,
"sd": 0.18
},
"p95": {
"mean": 23.17,
"sd": 0.26
},
"p99": {
"mean": 25.11,
"sd": 0.63
},
"mean": {
"mean": 21.77,
"sd": 0.17
}
},
"throughput_decisions_per_s": 45.93,
"cost_usd_per_1k_decisions": 0.0121,
"load_seconds": 9.836666666666666,
"rss_peak_mib": 2795.09375,
"gpu_mem_peak_mib": 1811.0,
"storage_mib": {
"weights": 647,
"environment": 4109
},
"prediction_agreement_across_reps": 1.0,
"top_confusions": [
{
"gold": "sem_correspondencia",
"pred": "cancelamento_linha",
"n": 26
},
{
"gold": "suporte_sem_sinal",
"pred": "internet_problema_conexao",
"n": 25
},
{
"gold": "sem_correspondencia",
"pred": "internet_problema_conexao",
"n": 23
},
{
"gold": "sem_correspondencia",
"pred": "vas_cancelamento",
"n": 15
},
{
"gold": "sem_correspondencia",
"pred": "internet_consumo",
"n": 14
},
{
"gold": "sem_correspondencia",
"pred": "reclamacao_atendimento",
"n": 14
},
{
"gold": "sem_correspondencia",
"pred": "atendimento_humano",
"n": 14
},
{
"gold": "sem_correspondencia",
"pred": "seguranca_fraude",
"n": 13
},
{
"gold": "seguranca_perda_roubo",
"pred": "seguranca_fraude",
"n": 10
},
{
"gold": "portabilidade",
"pred": "troca_numero",
"n": 10
}
],
"error_classes": {
"taxonomia_par_dificil": 182,
"fora_de_escopo_aceito": 140,
"intencao_fora_das_opcoes": 128,
"desempenho_do_modelo": 90,
"linguagem": 82,
"ambiguidade": 71,
"abstencao_indevida": 61
},
"slices": {
"profile": {
"curto_neutro": {
"n": 251,
"accuracy": 0.6016
},
"emocional": {
"n": 309,
"accuracy": 0.5858
},
"erros_digitacao": {
"n": 228,
"accuracy": 0.5702
},
"indireto_multiplo": {
"n": 239,
"accuracy": 0.5314
},
"informal_abreviado": {
"n": 216,
"accuracy": 0.5926
},
"longo_contexto": {
"n": 251,
"accuracy": 0.5538
},
"regionalismo": {
"n": 250,
"accuracy": 0.536
}
},
"n_options": {
"12": {
"n": 422,
"accuracy": 0.5142
},
"16": {
"n": 440,
"accuracy": 0.4795
},
"4": {
"n": 451,
"accuracy": 0.694
},
"8": {
"n": 431,
"accuracy": 0.5777
}
},
"kind": {
"intent": {
"n": 1508,
"accuracy": 0.5942
},
"oos": {
"n": 236,
"accuracy": 0.3983
}
},
"gold_eligible": {
"False": {
"n": 142,
"accuracy": 0.0915
},
"True": {
"n": 1602,
"accuracy": 0.6099
}
},
"length_bucket": {
"curto": {
"n": 578,
"accuracy": 0.5882
},
"longo": {
"n": 109,
"accuracy": 0.5413
},
"medio": {
"n": 1057,
"accuracy": 0.5591
}
},
"sentiment": {
"ansioso": {
"n": 330,
"accuracy": 0.5424
},
"confuso": {
"n": 3,
"accuracy": 0.6667
},
"frustrado": {
"n": 340,
"accuracy": 0.5441
},
"indignado": {
"n": 122,
"accuracy": 0.5574
},
"irritado": {
"n": 243,
"accuracy": 0.5226
},
"neutro": {
"n": 640,
"accuracy": 0.6078
},
"positivo": {
"n": 66,
"accuracy": 0.6061
}
},
"domain": {
"atendimento": {
"n": 72,
"accuracy": 0.5417
},
"beneficios_fidelidade": {
"n": 179,
"accuracy": 0.6536
},
"contas": {
"n": 219,
"accuracy": 0.5525
},
"fora_de_escopo": {
"n": 236,
"accuracy": 0.4025
},
"internet": {
"n": 108,
"accuracy": 0.7222
},
"linha_cadastro": {
"n": 145,
"accuracy": 0.6966
},
"marketplace": {
"n": 107,
"accuracy": 0.5607
},
"negociacao": {
"n": 72,
"accuracy": 0.375
},
"plano": {
"n": 108,
"accuracy": 0.6111
},
"recarga": {
"n": 108,
"accuracy": 0.7037
},
"roaming": {
"n": 108,
"accuracy": 0.6574
},
"seguranca": {
"n": 71,
"accuracy": 0.2958
},
"servicos_adicionais": {
"n": 104,
"accuracy": 0.5288
},
"suporte": {
"n": 107,
"accuracy": 0.5888
}
},
"annotation": {
"consenso": {
"n": 1723,
"accuracy": 0.567
},
"desempate": {
"n": 21,
"accuracy": 0.619
}
}
}
}
},
"rizzo_flow": {
"cuda": {
"device": "cuda",
"precision": "gguf-q8_0",
"repetitions": 3,
"quality": {
"n": 1744,
"errors": 0,
"accuracy": 0.8544,
"accuracy_ci95": [
0.8324,
0.8762
],
"macro_f1": 0.8771,
"macro_precision": 0.8584,
"macro_recall": 0.9123,
"fallback_recall": 0.6243,
"fallback_precision": 0.9593
},
"calibration": {
"temperature": 1.2523,
"raw": {
"brier": 0.2281,
"nll": 0.5092,
"ece": 0.0371
},
"calibrated": {
"brier": 0.2256,
"nll": 0.4926,
"ece": 0.0225
}
},
"abstention": {
"threshold": 0.923792,
"coverage": 0.4679,
"selective_accuracy": 0.9669,
"confident_wrong_rate": 0.0155
},
"latency_ms": {
"p50": {
"mean": 114.68,
"sd": 0.07
},
"p95": {
"mean": 180.82,
"sd": 0.43
},
"p99": {
"mean": 185.35,
"sd": 0.23
},
"mean": {
"mean": 125.95,
"sd": 0.31
}
},
"throughput_decisions_per_s": 7.94,
"cost_usd_per_1k_decisions": 0.07,
"load_seconds": 5.433333333333334,
"rss_peak_mib": 4534.83984375,
"gpu_mem_peak_mib": 5115.0,
"storage_mib": {
"weights": 4172,
"environment": 661
},
"prediction_agreement_across_reps": 1.0,
"top_confusions": [
{
"gold": "sem_correspondencia",
"pred": "conta_contestacao",
"n": 17
},
{
"gold": "sem_correspondencia",
"pred": "reclamacao_atendimento",
"n": 14
},
{
"gold": "sem_correspondencia",
"pred": "internet_problema_conexao",
"n": 8
},
{
"gold": "vas_cancelamento",
"pred": "conta_contestacao",
"n": 8
},
{
"gold": "conta_explicacao",
"pred": "conta_contestacao",
"n": 7
},
{
"gold": "sem_correspondencia",
"pred": "roaming_duvida",
"n": 6
},
{
"gold": "atendimento_humano",
"pred": "reclamacao_atendimento",
"n": 6
},
{
"gold": "sem_correspondencia",
"pred": "marketplace_pedido",
"n": 6
},
{
"gold": "sem_correspondencia",
"pred": "plano_nova_linha",
"n": 5
},
{
"gold": "sem_correspondencia",
"pred": "vas_contratacao",
"n": 5
}
],
"error_classes": {
"intencao_fora_das_opcoes": 106,
"taxonomia_par_dificil": 69,
"fora_de_escopo_aceito": 35,
"ambiguidade": 18,
"abstencao_indevida": 10,
"desempenho_do_modelo": 8,
"linguagem": 8
},
"slices": {
"profile": {
"curto_neutro": {
"n": 251,
"accuracy": 0.8924
},
"emocional": {
"n": 309,
"accuracy": 0.8544
},
"erros_digitacao": {
"n": 228,
"accuracy": 0.8465
},
"indireto_multiplo": {
"n": 239,
"accuracy": 0.7866
},
"informal_abreviado": {
"n": 216,
"accuracy": 0.8519
},
"longo_contexto": {
"n": 251,
"accuracy": 0.8685
},
"regionalismo": {
"n": 250,
"accuracy": 0.876
}
},
"n_options": {
"12": {
"n": 422,
"accuracy": 0.8341
},
"16": {
"n": 440,
"accuracy": 0.7909
},
"4": {
"n": 451,
"accuracy": 0.9024
},
"8": {
"n": 431,
"accuracy": 0.8886
}
},
"kind": {
"intent": {
"n": 1508,
"accuracy": 0.8554
},
"oos": {
"n": 236,
"accuracy": 0.8475
}
},
"gold_eligible": {
"False": {
"n": 142,
"accuracy": 0.2465
},
"True": {
"n": 1602,
"accuracy": 0.9082
}
},
"length_bucket": {
"curto": {
"n": 578,
"accuracy": 0.8772
},
"longo": {
"n": 109,
"accuracy": 0.8899
},
"medio": {
"n": 1057,
"accuracy": 0.8382
}
},
"sentiment": {
"ansioso": {
"n": 330,
"accuracy": 0.8515
},
"confuso": {
"n": 3,
"accuracy": 0.6667
},
"frustrado": {
"n": 340,
"accuracy": 0.8294
},
"indignado": {
"n": 122,
"accuracy": 0.8361
},
"irritado": {
"n": 243,
"accuracy": 0.8189
},
"neutro": {
"n": 640,
"accuracy": 0.8828
},
"positivo": {
"n": 66,
"accuracy": 0.8939
}
},
"domain": {
"atendimento": {
"n": 72,
"accuracy": 0.7222
},
"beneficios_fidelidade": {
"n": 179,
"accuracy": 0.8324
},
"contas": {
"n": 219,
"accuracy": 0.863
},
"fora_de_escopo": {
"n": 236,
"accuracy": 0.8517
},
"internet": {
"n": 108,
"accuracy": 0.8796
},
"linha_cadastro": {
"n": 145,
"accuracy": 0.8759
},
"marketplace": {
"n": 107,
"accuracy": 0.8879
},
"negociacao": {
"n": 72,
"accuracy": 0.9028
},
"plano": {
"n": 108,
"accuracy": 0.8148
},
"recarga": {
"n": 108,
"accuracy": 0.8704
},
"roaming": {
"n": 108,
"accuracy": 0.8241
},
"seguranca": {
"n": 71,
"accuracy": 0.9014
},
"servicos_adicionais": {
"n": 104,
"accuracy": 0.8173
},
"suporte": {
"n": 107,
"accuracy": 0.9065
}
},
"annotation": {
"consenso": {
"n": 1723,
"accuracy": 0.8561
},
"desempate": {
"n": 21,
"accuracy": 0.7143
}
}
}
}
},
"semif": {
"cuda": {
"device": "cuda",
"precision": "bfloat16",
"repetitions": 3,
"quality": {
"n": 1744,
"errors": 0,
"accuracy": 0.8721,
"accuracy_ci95": [
0.8495,
0.8941
],
"macro_f1": 0.8889,
"macro_precision": 0.893,
"macro_recall": 0.897,
"fallback_recall": 0.7672,
"fallback_precision": 0.8555
},
"calibration": {
"temperature": 1.284,
"raw": {
"brier": 0.1921,
"nll": 0.4137,
"ece": 0.0416
},
"calibrated": {
"brier": 0.1885,
"nll": 0.3976,
"ece": 0.0154
}
},
"abstention": {
"threshold": 0.914887,
"coverage": 0.6216,
"selective_accuracy": 0.9742,
"confident_wrong_rate": 0.0161
},
"latency_ms": {
"p50": {
"mean": 120.7,
"sd": 1.05
},
"p95": {
"mean": 187.12,
"sd": 1.75
},
"p99": {
"mean": 189.8,
"sd": 1.62
},
"mean": {
"mean": 128.55,
"sd": 1.09
}
},
"throughput_decisions_per_s": 7.78,
"cost_usd_per_1k_decisions": 0.0714,
"load_seconds": 7.34,
"rss_peak_mib": 9028.6875,
"gpu_mem_peak_mib": 8513.0,
"storage_mib": {
"weights": 8910,
"environment": 4113
},
"prediction_agreement_across_reps": 1.0,
"top_confusions": [
{
"gold": "roaming_cobranca",
"pred": "conta_contestacao",
"n": 14
},
{
"gold": "sem_correspondencia",
"pred": "conta_contestacao",
"n": 9
},
{
"gold": "sem_correspondencia",
"pred": "recarga_saldo",
"n": 7
},
{
"gold": "marketplace_compra",
"pred": "sem_correspondencia",
"n": 7
},
{
"gold": "sem_correspondencia",
"pred": "roaming_duvida",
"n": 7
},
{
"gold": "sem_correspondencia",
"pred": "cadastro_atualizacao",
"n": 7
},
{
"gold": "sem_correspondencia",
"pred": "suporte_sem_sinal",
"n": 5
},
{
"gold": "vas_cancelamento",
"pred": "conta_contestacao",
"n": 5
},
{
"gold": "sem_correspondencia",
"pred": "reclamacao_atendimento",
"n": 5
},
{
"gold": "plano_consulta",
"pred": "internet_consumo",
"n": 4
}
],
"error_classes": {
"intencao_fora_das_opcoes": 84,
"taxonomia_par_dificil": 62,
"abstencao_indevida": 49,
"ambiguidade": 17,
"desempenho_do_modelo": 4,
"fora_de_escopo_aceito": 4,
"linguagem": 3
},
"slices": {
"profile": {
"curto_neutro": {
"n": 251,
"accuracy": 0.9044
},
"emocional": {
"n": 309,
"accuracy": 0.8576
},
"erros_digitacao": {
"n": 228,
"accuracy": 0.886
},
"indireto_multiplo": {
"n": 239,
"accuracy": 0.7866
},
"informal_abreviado": {
"n": 216,
"accuracy": 0.8657
},
"longo_contexto": {
"n": 251,
"accuracy": 0.8685
},
"regionalismo": {
"n": 250,
"accuracy": 0.936
}
},
"n_options": {
"12": {
"n": 422,
"accuracy": 0.8578
},
"16": {
"n": 440,
"accuracy": 0.8523
},
"4": {
"n": 451,
"accuracy": 0.8869
},
"8": {
"n": 431,
"accuracy": 0.891
}
},
"kind": {
"intent": {
"n": 1508,
"accuracy": 0.8561
},
"oos": {
"n": 236,
"accuracy": 0.9746
}
},
"gold_eligible": {
"False": {
"n": 142,
"accuracy": 0.4085
},
"True": {
"n": 1602,
"accuracy": 0.9132
}
},
"length_bucket": {
"curto": {
"n": 578,
"accuracy": 0.91
},
"longo": {
"n": 109,
"accuracy": 0.8807
},
"medio": {
"n": 1057,
"accuracy": 0.8505
}
},
"sentiment": {
"ansioso": {
"n": 330,
"accuracy": 0.8576
},
"confuso": {
"n": 3,
"accuracy": 1.0
},
"frustrado": {
"n": 340,
"accuracy": 0.8794
},
"indignado": {
"n": 122,
"accuracy": 0.877
},
"irritado": {
"n": 243,
"accuracy": 0.8519
},
"neutro": {
"n": 640,
"accuracy": 0.8797
},
"positivo": {
"n": 66,
"accuracy": 0.8939
}
},
"domain": {
"atendimento": {
"n": 72,
"accuracy": 0.8194
},
"beneficios_fidelidade": {
"n": 179,
"accuracy": 0.838
},
"contas": {
"n": 219,
"accuracy": 0.9041
},
"fora_de_escopo": {
"n": 236,
"accuracy": 0.9831
},
"internet": {
"n": 108,
"accuracy": 0.9074
},
"linha_cadastro": {
"n": 145,
"accuracy": 0.8828
},
"marketplace": {
"n": 107,
"accuracy": 0.8505
},
"negociacao": {
"n": 72,
"accuracy": 0.8611
},
"plano": {
"n": 108,
"accuracy": 0.8056
},
"recarga": {
"n": 108,
"accuracy": 0.8611
},
"roaming": {
"n": 108,
"accuracy": 0.7315
},
"seguranca": {
"n": 71,
"accuracy": 0.9014
},
"servicos_adicionais": {
"n": 104,
"accuracy": 0.7981
},
"suporte": {
"n": 107,
"accuracy": 0.9065
}
},
"annotation": {
"consenso": {
"n": 1723,
"accuracy": 0.8752
},
"desempate": {
"n": 21,
"accuracy": 0.619
}
}
}
}
}
},
"pairwise_vs_best_gpu": {
"best": "semif",
"comparisons": {
"gliner": {
"accuracy_diff": 0.3022,
"ci95": [
0.2607,
0.3427
],
"a_only": 603,
"b_only": 76,
"p_value": 1.0283351885027499e-102
},
"laya": {
"accuracy_diff": 0.3045,
"ci95": [
0.2653,
0.3464
],
"a_only": 605,
"b_only": 74,
"p_value": 1.5972799301464988e-104
},
"rizzo_flow": {
"accuracy_diff": 0.0178,
"ci95": [
-0.0052,
0.0418
],
"a_only": 127,
"b_only": 96,
"p_value": 0.044300169930155384
}
}
},
"criteria_check": {
"gliner": {
"H3_latency_p95_gpu": true,
"H1_macro_f1_absolute": false,
"H2_risk": false
},
"laya": {
"H3_latency_p95_gpu": true,
"H1_macro_f1_absolute": false,
"H2_risk": false
},
"rizzo_flow": {
"H3_latency_p95_gpu": false,
"H1_macro_f1_absolute": true,
"H2_risk": false
},
"semif": {
"H3_latency_p95_gpu": false,
"H1_macro_f1_absolute": true,
"H2_risk": false
}
}
}