Decider-4B / evaluation /report.json
roudone's picture
Update typed-decisions evaluation
a9a6a79 verified
Raw History Blame Contribute Delete
6.71 kB
{
"dataset": "LocalLLaMA/typed-decisions",
"split": "test",
"selected_cases": 400,
"successful_cases": 400,
"failed_cases": 0,
"scored_cases": 400,
"overall": {
"decisions": 2000,
"correct": 1652,
"accuracy": 0.826,
"kl": 0.0803057449071361,
"brier": 0.04295575756754636,
"ece_15": 0.15975580646719287
},
"metric_notes": {
"accuracy": "argmax prediction against the gold argmax label",
"kl": "mean KL(gold || prediction), natural logarithm",
"brier": "mean sum over classes of (prediction - soft gold)^2",
"ece_15": "15 equal-width bins using top probability and hard argmax correctness"
},
"latency_ms": {
"p50": 106.56309803016484,
"p95": 135.15076198382303
},
"scoring_errors": [],
"by_workflow": {
"agent_trace_observability": {
"decisions": 500,
"correct": 389,
"accuracy": 0.778,
"kl": 0.08059402466036933,
"brier": 0.0463716026092487,
"ece_15": 0.185638753939392
},
"customer_service": {
"decisions": 500,
"correct": 425,
"accuracy": 0.85,
"kl": 0.08023071358764519,
"brier": 0.04071135055898889,
"ece_15": 0.15487085697191613
},
"invoice_processing": {
"decisions": 500,
"correct": 429,
"accuracy": 0.858,
"kl": 0.09564322592504344,
"brier": 0.05099897025345711,
"ece_15": 0.09557565599617864
},
"security_incidents": {
"decisions": 500,
"correct": 409,
"accuracy": 0.818,
"kl": 0.06475501545548643,
"brier": 0.033741106848490764,
"ece_15": 0.20823487398337517
}
},
"by_type": {
"choice": {
"decisions": 600,
"correct": 478,
"accuracy": 0.7966666666666666,
"kl": 0.10667343256268746,
"brier": 0.05479604004707588,
"ece_15": 0.15765588434266006
},
"noul": {
"decisions": 600,
"correct": 527,
"accuracy": 0.8783333333333333,
"kl": 0.04097212613378347,
"brier": 0.030015148967631963,
"ece_15": 0.10459395721244316
},
"score": {
"decisions": 800,
"correct": 647,
"accuracy": 0.80875,
"kl": 0.09003019324548704,
"brier": 0.04378100215783503,
"ece_15": 0.20270213500165474
}
},
"by_question": {
"agent_trace_observability/action": {
"decisions": 100,
"correct": 72,
"accuracy": 0.72,
"kl": 0.08790775035944622,
"brier": 0.045996883503236996,
"ece_15": 0.1923458341441178
},
"agent_trace_observability/needs_review": {
"decisions": 100,
"correct": 90,
"accuracy": 0.9,
"kl": 0.033611987181820735,
"brier": 0.02631921507244329,
"ece_15": 0.17988898418843746
},
"agent_trace_observability/outcome": {
"decisions": 100,
"correct": 75,
"accuracy": 0.75,
"kl": 0.10884891853683683,
"brier": 0.06128768621134422,
"ece_15": 0.22472407297158692
},
"agent_trace_observability/risk": {
"decisions": 100,
"correct": 82,
"accuracy": 0.82,
"kl": 0.08616333697498625,
"brier": 0.051944564727536655,
"ece_15": 0.21548420564294987
},
"agent_trace_observability/urgency": {
"decisions": 100,
"correct": 70,
"accuracy": 0.7,
"kl": 0.0864381302487566,
"brier": 0.04630966353168234,
"ece_15": 0.17894596456783354
},
"customer_service/action": {
"decisions": 100,
"correct": 72,
"accuracy": 0.72,
"kl": 0.16485836116812386,
"brier": 0.08073784085652039,
"ece_15": 0.1081227785754724
},
"customer_service/category": {
"decisions": 100,
"correct": 98,
"accuracy": 0.98,
"kl": 0.04316980921110677,
"brier": 0.015287821774841741,
"ece_15": 0.12546654650846617
},
"customer_service/churn_risk": {
"decisions": 100,
"correct": 89,
"accuracy": 0.89,
"kl": 0.06087967761779349,
"brier": 0.02739846473868238,
"ece_15": 0.24890314749550854
},
"customer_service/needs_human": {
"decisions": 100,
"correct": 82,
"accuracy": 0.82,
"kl": 0.05379561473975391,
"brier": 0.0447702723712042,
"ece_15": 0.10668263003230095
},
"customer_service/urgency": {
"decisions": 100,
"correct": 84,
"accuracy": 0.84,
"kl": 0.07845010520144792,
"brier": 0.035362353053695735,
"ece_15": 0.23016179276997029
},
"invoice_processing/discrepancy_severity": {
"decisions": 100,
"correct": 86,
"accuracy": 0.86,
"kl": 0.16211923361755753,
"brier": 0.07577871556224147,
"ece_15": 0.11399502083683016
},
"invoice_processing/disposition": {
"decisions": 100,
"correct": 81,
"accuracy": 0.81,
"kl": 0.14687131365910766,
"brier": 0.08027354501411901,
"ece_15": 0.14137618761078863
},
"invoice_processing/duplicate": {
"decisions": 100,
"correct": 91,
"accuracy": 0.91,
"kl": 0.03952439425291494,
"brier": 0.026807131545151866,
"ece_15": 0.0636753043998033
},
"invoice_processing/matches_order": {
"decisions": 100,
"correct": 95,
"accuracy": 0.95,
"kl": 0.05999561126647583,
"brier": 0.03594590921788439,
"ece_15": 0.060309361563995516
},
"invoice_processing/urgency": {
"decisions": 100,
"correct": 76,
"accuracy": 0.76,
"kl": 0.06970557682916122,
"brier": 0.036189549927888774,
"ece_15": 0.17601861818019318
},
"security_incidents/credential_compromise": {
"decisions": 100,
"correct": 82,
"accuracy": 0.82,
"kl": 0.031176862771824396,
"brier": 0.025240369597254987,
"ece_15": 0.14930089343339203
},
"security_incidents/disposition": {
"decisions": 100,
"correct": 80,
"accuracy": 0.8,
"kl": 0.0883844424415034,
"brier": 0.04519246292239289,
"ece_15": 0.2498926800447473
},
"security_incidents/severity": {
"decisions": 100,
"correct": 84,
"accuracy": 0.84,
"kl": 0.07871596952034915,
"brier": 0.0302813399315595,
"ece_15": 0.3170699022538294
},
"security_incidents/true_positive": {
"decisions": 100,
"correct": 87,
"accuracy": 0.87,
"kl": 0.027728286589911,
"brier": 0.02100799600185304,
"ece_15": 0.19495224751532075
},
"security_incidents/urgency": {
"decisions": 100,
"correct": 76,
"accuracy": 0.76,
"kl": 0.09776951595384421,
"brier": 0.0469833657893934,
"ece_15": 0.2065293719543915
}
}
}