Files
decision-models-benchmark/reports/pilot/summary-laya-cuda.json
2026-10-07 18:11:45 -03:00

61 lines
1.3 KiB
JSON

{
"candidate": {
"name": "laya",
"model_id": "laya@1720e3e3357c",
"revision": "1720e3e3357cfe1e281542e223f8273b0890ca34",
"adapter_version": "0.2.0",
"device": "cuda",
"context_limit": 1024,
"probabilities_source": "native"
},
"precision": "float32",
"device": "cuda",
"split": "dev",
"examples": 50,
"seed": 20261007,
"warmup": 5,
"batch_size": 1,
"state_budget_tokens": 512,
"process_guard": "ok",
"load_seconds": 9.06,
"warmup_latency_ms": [
264.4,
22.4,
28.5,
22.4,
22.5
],
"latency_ms": {
"n": 45,
"p50": 22.042784999996456,
"p95": 22.97260880000067,
"p99": 23.23121632001289,
"mean": 21.861335755556233
},
"errors": {},
"truncated": 0,
"with_diagnostics": 0,
"operational_env": {},
"state_tokens_max": 88,
"rss_peak_mib": 2773.41015625,
"gpu_mem_peak_mib": 1811.0,
"gpu_util_mean_pct": 6.857142857142857,
"quality": {
"n": 50,
"errors": 0,
"accuracy": 0.66,
"macro_f1": 0.6346938775510205,
"macro_precision": 0.6244047619047619,
"macro_recall": 0.7142857142857142
},
"calibration_raw": {
"brier": 0.5653510164045072,
"nll": 2.310904764203164,
"ece": 0.2660502261031441
},
"oci": {
"region": "sa-saopaulo-1",
"run_id": "20261007T180950Z-laya-evaluate"
}
}