SUBMISSION / VERIFIED
Dampen Wait/Hmm token embeddings (scale 0.4)
Verified
Starting evaluation…
VALIDATION41.36%Equal-weight average of public benchmarks-0.47 points vs. original
PRIVATE TEST22.40%Determines leaderboard rank3.80 points vs. original
Validation breakdown
| Public benchmark | Score |
|---|---|
| bbh | 37.04% |
| mbpp | 18.75% |
| mmlu | 46.49% |
| piqa | 71.88% |
| boolq | 64.06% |
| gsm8k | 53.13% |
| ifeval | 53.13% |
| math500 | 46.88% |
| arc_easy | 64.06% |
| mmlu_pro | 27.14% |
| hellaswag | 46.88% |
| humaneval | 25.00% |
| openbookqa | 29.69% |
| winogrande | 56.25% |
| arc_challenge | 32.81% |
| commonsense_qa | 54.69% |
| bfcl_multi_turn | 0.00% |
| bfcl_single_turn | 16.67% |
Evaluation history
Earlier scores remain visible when the test set changes. Compare scores only within the same evaluation version.
| Completed | Validation | Private test |
|---|---|---|
| Nov 5, 2026 · 05:39 UTC | 41.36% | 22.40% |
A dash means no result yet. Private test questions, answers, and individual results are never included here.