Entrant: @spartanhaden.
Three sandboxed Modal A100-80GB runs of the unchanged scorer (mnist-a100/1.2.0, run_modal.py --difficulty 2 --runs 3) all passed.
| Run | GPU | MNIST error | Margin (labels) | Worst draw | Hold-out | Ranked ms | mJ above idle |
|---|---|---|---|---|---|---|---|
| 1 | A100-SXM4-80GB | 3.069% | +364 | 3.23% | Fashion 87.50% | 4.533 | 570 |
| 2 | A100-SXM4-80GB | 2.997% | +443 | 3.31% | Fashion 87.53% | 5.092 | 601 |
| 3 | A100-SXM4-80GB | 3.133% | +294 | 3.32% | Fashion 87.11% | 4.736 | 370 |
Compared with the current record (14.6 ms, 1,333 mJ; @sjbaebae, native MLPs with three-neighbor voting), this entry is 3.1× faster.
submission.py is 12,480 bytes with SHA-256 d2e38e1072b2f7ca6e0e3ade3746d2bddb32604aca8d1ddcc6127233135f2776, identical to source_sha256 in all three records. It passes check_source. evidence/official/manifest.json lists the hash of every file.
Builds directly on @islamborghini’s kernel-pcg entry (RBF kernel ridge, one AGOP metric update, Nyström-preconditioned CG). This file re-tunes it and rebuilds it in plain PyTorch. The AGOP metric step is from Recursive Feature Machines (Radhakrishnan et al. 2022, arXiv 2212.13881).
torch.compile, so the warm-up is short and the scorer’s post-warm-up staging cost stays small.All three initial fresh energy-enabled sandboxed Modal A100-80GB runs passed individually. Source review found no qualification blocker. All 45 timed calls and complete energy windows were reconstructed with the unchanged canonical scorer; three distinct GPU UUIDs were recorded. No failed run was replaced or learner repair made.
The reviewer medians are 4.578529 ms and 602.662112 mJ above idle per call. The leaderboard uses this independent set; the author source and original evidence above are retained.
| Reviewer run | Correct | MNIST error | Ranked ms | Above-idle mJ |
|---|---|---|---|---|
| 1 | 106,444/110,000 | 3.232727% | 4.382246 | 602.662112 |
| 2 | 106,529/110,000 | 3.155455% | 4.578529 | 628.332434 |
| 3 | 106,558/110,000 | 3.129091% | 5.892114 | 425.005391 |
Raw records and manifest and canonical reconstruction are retained. Learner SHA-256: d2e38e1072b2f7ca6e0e3ade3746d2bddb32604aca8d1ddcc6127233135f2776. Energy excludes host energy and first-call compilation.
Reproduce from mnist-a100/:
python run_modal.py submissions/rfm-pcg-d2-20261012/submission.py:classify --difficulty 2 --runs 3 --json /tmp/pr121-d2-verification