openjev-e4b / eval /openjev_1_0_public_benchmark.json
bambamdevs's picture
Publish OpenJEV E4B 1.0
03223d7
Raw History Blame Contribute Delete
2.26 kB
{
"model_line": "openjev-e4b",
"release": "1.0",
"benchmark_artifact": "local NF4 frozen base + dense adapted layers + LoRA + decision head",
"source_summary": "eval/raw_run/summary.json",
"raw_control_summary": "eval/raw_e4b_nf4_control_summary.json",
"tasks": {
"arc_challenge": {
"status": "measured",
"display": "ARC-Challenge",
"n": 1172,
"accuracy": 0.878839590443686,
"nll": 0.3434354264088279,
"brier": 0.18004018282065612,
"ece": 0.018640702693919233,
"p50_ms": 173.45179200128769,
"p95_ms": 241.31601209974178,
"raw_e4b_nf4_accuracy": 0.8737,
"delta_vs_raw_pp": 0.5139590443685949
},
"winogrande": {
"status": "measured",
"display": "WinoGrande",
"n": 1267,
"accuracy": 0.664561957379637,
"nll": 0.8732282598352692,
"brier": 0.5079363220206212,
"ece": 0.1986676781891346,
"p50_ms": 168.07529400102794,
"p95_ms": 241.4132092009823,
"raw_e4b_nf4_accuracy": 0.6014,
"delta_vs_raw_pp": 6.316195737963692
},
"arc_easy": {
"status": "measured",
"display": "ARC-Easy",
"n": 2376,
"accuracy": 0.9562289562289562,
"nll": 0.15331558248276225,
"brier": 0.07269179989034444,
"ece": 0.029571857402955024,
"p50_ms": 172.40582649901626,
"p95_ms": 231.58336125015921,
"raw_e4b_nf4_accuracy": 0.9558,
"delta_vs_raw_pp": 0.04289562289562232
},
"hellaswag": {
"status": "measured",
"display": "HellaSwag",
"n": 10042,
"accuracy": 0.8977295359490142,
"nll": 0.2736215284007136,
"brier": 0.14571887516017584,
"ece": 0.012820075397336044,
"p50_ms": 230.12878749796073,
"p95_ms": 276.0439652007335,
"raw_e4b_nf4_accuracy": 0.7651,
"delta_vs_raw_pp": 13.262953594901417
},
"mmlu": {
"status": "measured",
"display": "MMLU",
"n": 14042,
"accuracy": 0.6441390115368181,
"nll": 0.8705675106095903,
"brier": 0.45396635565340543,
"ece": 0.03314363695835977,
"p50_ms": 196.04511249963252,
"p95_ms": 303.28189720094076,
"raw_e4b_nf4_accuracy": 0.6619,
"delta_vs_raw_pp": -1.776098846318197
}
}
}