mirror of
https://github.com/mcloh/decision-models-benchmark.git
synced 2026-11-05 22:04:12 +00:00
59 lines
1.3 KiB
JSON
59 lines
1.3 KiB
JSON
{
|
|
"candidate": {
|
|
"name": "rizzo_flow",
|
|
"model_id": "rizzo_flow@55633c8cbd2b",
|
|
"revision": "55633c8cbd2b826bd3eefdeb05310450996649df",
|
|
"adapter_version": "0.1.0",
|
|
"device": "cuda",
|
|
"context_limit": 2048,
|
|
"probabilities_source": "native"
|
|
},
|
|
"precision": "gguf-q8_0",
|
|
"device": "cuda",
|
|
"split": "dev",
|
|
"examples": 50,
|
|
"seed": 20261007,
|
|
"warmup": 5,
|
|
"batch_size": 1,
|
|
"state_budget_tokens": 512,
|
|
"process_guard": "ok",
|
|
"load_seconds": 5.34,
|
|
"warmup_latency_ms": [
|
|
132.8,
|
|
103.2,
|
|
80.3,
|
|
75.0,
|
|
101.8
|
|
],
|
|
"latency_ms": {
|
|
"n": 45,
|
|
"p50": 108.08683399997676,
|
|
"p95": 174.65078400000493,
|
|
"p99": 182.52244964000855,
|
|
"mean": 121.8410705555527
|
|
},
|
|
"errors": {},
|
|
"truncated": 0,
|
|
"state_tokens_max": 118,
|
|
"rss_peak_mib": 4527.43359375,
|
|
"gpu_mem_peak_mib": 5107.0,
|
|
"gpu_util_mean_pct": 40.13636363636363,
|
|
"quality": {
|
|
"n": 50,
|
|
"errors": 0,
|
|
"accuracy": 0.82,
|
|
"macro_f1": 0.8428571428571429,
|
|
"macro_precision": 0.8373015873015872,
|
|
"macro_recall": 0.8839285714285714
|
|
},
|
|
"calibration_raw": {
|
|
"brier": 0.23497548126089615,
|
|
"nll": 0.44902628720146837,
|
|
"ece": 0.10324444394049462
|
|
},
|
|
"oci": {
|
|
"region": "sa-saopaulo-1",
|
|
"run_id": "20261007T180058Z-rizzo_flow-evaluate"
|
|
}
|
|
}
|