evals

Vidya

completestandardruns/polar-dense: polar dense 4L/32H, 30000 updates (983 MB), 1.8469 bpb, 3.62 M MAC/byte, 2.14e+16 FLOP

CORE 0.0182(+0.0093 vs previous)

2026-10-21 20:36 · max 500 examples/task · scorer v1 · bundle abc150c04fa3 · endpoint https://core.roland.blah.dev

TaskAccuracyCenteredΔ centerednTime
agi_eval_lsat_ar0.21300.0163-0.0815230443s
arc_challenge0.25800.0107-0.0213500979s
arc_easy0.31400.0853+0.0027500823s
bigbench_cs_algorithms0.00000.00000.0000500162s
bigbench_dyck_languages0.05600.0560+0.0220500194s
bigbench_language_identification0.23800.1617+0.0022500976s
bigbench_operators0.07620.0762+0.009521066s
bigbench_qa_wikidata0.00000.00000.000050065s
bigbench_repeat_copy_logic0.00000.00000.00003212s
boolq0.4680-0.4000+0.2526500511s
commonsense_qa0.42000.2750+0.0200500794s
copa0.53000.0600-0.06001008s
coqa0.00000.00000.0000500215s
hellaswag0.25200.0027+0.0240500970s
hellaswag_zeroshot0.2440-0.0080+0.0107500188s
jeopardy0.00000.00000.0000500149s
lambada_openai0.00000.00000.000050050s
openbook_qa0.2460-0.0053-0.018750052s
piqa0.50600.0120+0.0440500394s
squad0.00000.00000.0000500257s
winograd0.51280.0256-0.014727325s
winogrande0.51600.0320+0.012050046s

Every example's outcome (per-option losses, greedy matches, correctness) is stored with this attempt · deltas are against the previous standard attempt.