QEV2-12B / evaluation.json
querypieai's picture
Add merged inference-only model, multilingual cards and research license
36b8f8e verified
Raw History Blame Contribute Delete
645 Bytes
{
"historical_metrics": {
"mmlu": 0.4368896484375,
"belebele": 0.688936723255625,
"private": 0.9185185185185185,
"composite": 0.6814482967372145,
"dlp_accuracy": 0.9829491632459741,
"dlp_recall": 0.9968774395003903,
"dlp_f2": 0.9897690280576655,
"dlp_fp": 50,
"dlp_fn": 4
},
"new_full_benchmark": false,
"counts": {
"mmlu": 16384,
"belebele": 24622,
"private": 540,
"dlp": 3167
},
"dlp_threshold": 0.5,
"context": {
"state": 384,
"branch": 1024,
"packed": 2048
},
"scope": "Historical unmerged BF16 source only; NOT newly measured merged/INT8 benchmark scores"
}