openjev-e4b / eval /openjev_1_0_public_benchmark.csv
bambamdevs's picture
Publish OpenJEV E4B 1.0
03223d7
Raw History Blame Contribute Delete
980 Bytes
task,display,status,n,accuracy,nll,brier,ece,p50_ms,p95_ms,raw_e4b_nf4_accuracy,delta_vs_raw_pp
arc_challenge,ARC-Challenge,measured,1172,0.878839590443686,0.3434354264088279,0.18004018282065612,0.018640702693919233,173.45179200128769,241.31601209974178,0.8737,0.5139590443685949
winogrande,WinoGrande,measured,1267,0.664561957379637,0.8732282598352692,0.5079363220206212,0.1986676781891346,168.07529400102794,241.4132092009823,0.6014,6.316195737963692
arc_easy,ARC-Easy,measured,2376,0.9562289562289562,0.15331558248276225,0.07269179989034444,0.029571857402955024,172.40582649901626,231.58336125015921,0.9558,0.04289562289562232
hellaswag,HellaSwag,measured,10042,0.8977295359490142,0.2736215284007136,0.14571887516017584,0.012820075397336044,230.12878749796073,276.0439652007335,0.7651,13.262953594901417
mmlu,MMLU,measured,14042,0.6441390115368181,0.8705675106095903,0.45396635565340543,0.03314363695835977,196.04511249963252,303.28189720094076,0.6619,-1.776098846318197