Files
decision-models-benchmark/reports/dev/summary-gliner-cuda.json
2026-10-07 20:10:53 -03:00

62 lines
1.4 KiB
JSON

{
"candidate": {
"name": "gliner",
"model_id": "gliner@a35a0cd3b7a0",
"revision": "a35a0cd3b7a0f00f2effc576f454cd48fa98aa5f",
"adapter_version": "0.2.0",
"device": "cuda",
"context_limit": 4096,
"probabilities_source": "native"
},
"precision": "float32",
"device": "cuda",
"split": "dev",
"examples": 1183,
"seed": 20261007,
"warmup": 5,
"batch_size": 1,
"state_budget_tokens": 512,
"process_guard": "ok",
"repetition": 1,
"load_seconds": 21.83,
"warmup_latency_ms": [
230.5,
122.3,
102.2,
268.8,
32.8
],
"latency_ms": {
"n": 1178,
"p50": 18.382403499998645,
"p95": 31.400230849988017,
"p99": 33.525557409979854,
"mean": 21.355778809847063
},
"errors": {},
"truncated": 0,
"with_diagnostics": 0,
"operational_env": {},
"state_tokens_max": 111,
"rss_peak_mib": 3227.6171875,
"gpu_mem_peak_mib": 1959.0,
"gpu_util_mean_pct": 42.634408602150536,
"quality": {
"n": 1183,
"errors": 0,
"accuracy": 0.5545224006762468,
"macro_f1": 0.5525597734274297,
"macro_precision": 0.7029289830574256,
"macro_recall": 0.542247078905617
},
"calibration_raw": {
"brier": 0.6045309118425269,
"nll": 1.3080394222612395,
"ece": 0.08826073768581884
},
"oci": {
"region": "sa-saopaulo-1",
"run_id": "20261007T225945Z-gliner-evaluate"
}
}