QEV-9B / evaluation.json
querypieai's picture
Add merged inference-only model, multilingual cards and research license
96adfbe verified
Raw History Blame Contribute Delete
646 Bytes
{
"historical_metrics": {
"mmlu": 0.445068359375,
"belebele": 0.7328405491024287,
"private": 0.8796296296296297,
"composite": 0.6858461793690195,
"dlp_accuracy": 0.9851594568992738,
"dlp_recall": 0.9781420765027322,
"dlp_f2": 0.9795184490306442,
"dlp_fp": 19,
"dlp_fn": 28
},
"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"
}