mirror of
https://github.com/mcloh/decision-models-benchmark.git
synced 2026-11-05 22:04:12 +00:00
Piloto concluído (tarefas 52–58): reports/pilot, viabilidade em CPU e parâmetros mantidos
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
56
reports/pilot/summary-semif-cpu.json
Normal file
56
reports/pilot/summary-semif-cpu.json
Normal file
@@ -0,0 +1,56 @@
|
||||
{
|
||||
"candidate": {
|
||||
"name": "semif",
|
||||
"model_id": "semif@851bf6e806ef",
|
||||
"revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a",
|
||||
"adapter_version": "0.1.0",
|
||||
"device": "cpu",
|
||||
"context_limit": 8192,
|
||||
"probabilities_source": "native"
|
||||
},
|
||||
"precision": "bfloat16",
|
||||
"device": "cpu",
|
||||
"split": "dev",
|
||||
"examples": 50,
|
||||
"seed": 20261007,
|
||||
"warmup": 5,
|
||||
"batch_size": 1,
|
||||
"state_budget_tokens": 512,
|
||||
"process_guard": "ok",
|
||||
"load_seconds": 6.27,
|
||||
"warmup_latency_ms": [
|
||||
32460.9,
|
||||
28606.1,
|
||||
22025.6,
|
||||
20641.6,
|
||||
30299.5
|
||||
],
|
||||
"latency_ms": {
|
||||
"n": 45,
|
||||
"p50": 30633.44516099983,
|
||||
"p95": 47585.681381199945,
|
||||
"p99": 49343.981749680024,
|
||||
"mean": 33482.62709695556
|
||||
},
|
||||
"errors": {},
|
||||
"truncated": 0,
|
||||
"state_tokens_max": 89,
|
||||
"rss_peak_mib": 9136.75,
|
||||
"quality": {
|
||||
"n": 50,
|
||||
"errors": 0,
|
||||
"accuracy": 0.92,
|
||||
"macro_f1": 0.9392857142857143,
|
||||
"macro_precision": 0.9404761904761905,
|
||||
"macro_recall": 0.9464285714285714
|
||||
},
|
||||
"calibration_raw": {
|
||||
"brier": 0.1605347137797607,
|
||||
"nll": 0.3247898203379594,
|
||||
"ece": 0.0786511705708897
|
||||
},
|
||||
"oci": {
|
||||
"region": "us-chicago-1",
|
||||
"run_id": "20261007T174231Z-semif-evaluate"
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user