mirror of
https://github.com/mcloh/decision-models-benchmark.git
synced 2026-11-05 22:04:12 +00:00
62 lines
1.4 KiB
JSON
62 lines
1.4 KiB
JSON
{
|
|
"candidate": {
|
|
"name": "semif",
|
|
"model_id": "semif@851bf6e806ef",
|
|
"revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a",
|
|
"adapter_version": "0.2.0",
|
|
"device": "cuda",
|
|
"context_limit": 8192,
|
|
"probabilities_source": "native"
|
|
},
|
|
"precision": "bfloat16",
|
|
"device": "cuda",
|
|
"split": "test",
|
|
"examples": 1744,
|
|
"seed": 20261007,
|
|
"warmup": 5,
|
|
"batch_size": 1,
|
|
"state_budget_tokens": 512,
|
|
"process_guard": "ok",
|
|
"repetition": 3,
|
|
"load_seconds": 7.74,
|
|
"warmup_latency_ms": [
|
|
538.4,
|
|
106.2,
|
|
106.1,
|
|
74.8,
|
|
106.4
|
|
],
|
|
"latency_ms": {
|
|
"n": 1739,
|
|
"p50": 121.58077500004083,
|
|
"p95": 188.20931719997134,
|
|
"p99": 190.76599447997523,
|
|
"mean": 129.24359273145473
|
|
},
|
|
"errors": {},
|
|
"truncated": 0,
|
|
"with_diagnostics": 0,
|
|
"operational_env": {},
|
|
"state_tokens_max": 104,
|
|
"rss_peak_mib": 9028.3515625,
|
|
"gpu_mem_peak_mib": 8513.0,
|
|
"gpu_util_mean_pct": 88.56140350877193,
|
|
"quality": {
|
|
"n": 1744,
|
|
"errors": 0,
|
|
"accuracy": 0.8721330275229358,
|
|
"macro_f1": 0.8888570465419293,
|
|
"macro_precision": 0.8930113096096118,
|
|
"macro_recall": 0.8970062968520272
|
|
},
|
|
"calibration_raw": {
|
|
"brier": 0.19207800377371836,
|
|
"nll": 0.41373138162846335,
|
|
"ece": 0.04159838450005465
|
|
},
|
|
"oci": {
|
|
"region": "sa-saopaulo-1",
|
|
"run_id": "20261007T231653Z-semif-evaluate"
|
|
}
|
|
}
|