Escarda-86M

#11
by Quazim0t0 - opened

Model: https://huggingface.co/Quazim0t0/Escarda-86M

ArithMark-2.0 — Escarda-86M breakdown (acc_norm)
Overall: acc 0.2932 · acc_norm 0.2816 (n=2,500, chance = 0.25)

By difficulty:

Difficulty acc_norm n
easy 0.2872 1250
medium 0.2973 750
hard 0.2440 500

By topic:
Topic acc_norm n
division 0.5385 130
multiplication 0.5278 144
parentheses_two_ops 0.3352 355
mixed_two_ops 0.2633 395
parentheses_three_ops 0.2558 258
mixed_three_ops 0.2314 242
addition 0.2323 538
subtraction 0.2009 438

Normal Benchmarks:
arc_easy acc,none 0.3683

arc_easy acc_stderr,none 0.0099

arc_easy acc_norm,none 0.3628

arc_easy acc_norm_stderr,none 0.0099

arc_challenge acc,none 0.1988

arc_challenge acc_stderr,none 0.0117

arc_challenge acc_norm,none 0.2312

arc_challenge acc_norm_stderr,none 0.0123

hellaswag acc,none 0.2845

hellaswag acc_stderr,none 0.0045

hellaswag acc_norm,none 0.2928

hellaswag acc_norm_stderr,none 0.0045

winogrande acc,none 0.5067

winogrande acc_stderr,none 0.0140

piqa acc,none 0.5881

piqa acc_stderr,none 0.0115

piqa acc_norm,none 0.5800

piqa acc_norm_stderr,none 0.0115

openbookqa acc,none 0.1600

openbookqa acc_stderr,none 0.0164

openbookqa acc_norm,none 0.2720

openbookqa acc_norm_stderr,none 0.0199

boolq acc,none 0.4624

boolq acc_stderr,none 0.0087

Axiomic Labs org

Its up!

Datdanboi25 changed discussion status to closed

Sign up or log in to comment