laya-typed-decisions / laya_benchmark_report.json
Ankit1106's picture
Publish fine-tuned Laya derivative and official evaluation (accuracy 0.769)
66172d2 verified
Raw History Blame Contribute Delete
774 Bytes
{
"benchmark": "LocalLLaMA/typed-decisions",
"model": "Laya (Fine-Tuned 2xT4)",
"n_cases": 400,
"n_decisions": 2000,
"metrics": {
"accuracy": 0.769,
"soft_accuracy": 0.5068,
"brier_score": 0.0694,
"ece": 0.215,
"score_mae": 0.2442,
"within_1_level": 0.9912,
"latency_p50_ms": 116.1,
"latency_p95_ms": 153.8,
"kl_divergence": 0.1287,
"total_variation": 0.184
},
"per_workflow": {
"agent_trace_observability": {
"decisions": 500,
"accuracy": 0.746
},
"customer_service": {
"decisions": 500,
"accuracy": 0.776
},
"invoice_processing": {
"decisions": 500,
"accuracy": 0.806
},
"security_incidents": {
"decisions": 500,
"accuracy": 0.748
}
}
}