The first frozen reviewer group passed 3/3 fresh canonical sandboxed Modal A100-80GB runs, with valid above-idle energy measurements on three distinct GPUs. Median ranked time is 1,915.402496 ms/call and median energy is 343,151.622557 mJ/call; pooled MNIST error is 1.789091% (324,096/330,000 correct).
| Reviewer run | MNIST error | Margin, labels | Ranked ms | Above-idle mJ |
|---|---|---|---|---|
| 1 | 1.810000% | +99 | 1,922.228505 | 331,163.170874 |
| 2 | 1.812727% | +96 | 1,913.758972 | 343,151.622557 |
| 3 | 1.744545% | +171 | 1,915.402496 | 346,421.252166 |
Source is unchanged at SHA-256 9adc570a02279b34d0766344ec72f5589ff10fae06e1b6b341b58dc351a6445a. The scorer and runner are unchanged. Verification reconstructs every complete canonical score and energy result from the raw records; source review covers per-call fitting, label isolation and cached-state reset. No reviewer outcome was replaced or retried.
The leaderboard uses the independent reviewer medians above. Author results below are retained separately. The published instructions specify three official Modal runs and their median; this independent reproduction is review evidence, not a newly imposed second-batch submission rule.
Using Python 3.13 as in the canonical runner, recheck saved records from this submission directory with python evidence/reviewer-20261014/verify_records.py evidence/reviewer-20261014 submission.py --output /tmp/pr120-verification.json. This reconstructs recorded verdicts without executing the learner.
Present-tense record and pending-review claims below describe the submission date; the dated reviewer section above gives the accepted review status.
Entrant: @likelytobelaura. The joint train+query graph-propagation readout is @sjbaebae’s, from the D3 record (see Credits).
Three canonical Modal A100-80GB sandbox evaluations, the first and only set run for this source, all passed difficulty 5 (mean MNIST error at most 1.90%). Median ranked time is 1,898.373 ms per call; median energy above idle is 333,939.8 mJ per call. The 33 MNIST calls labelled 324,016 / 330,000 rows correctly (1.813% aggregate error). This is the same method as the current D5 record (tangent-crossteach-d5-20261012, 3,335.4 ms) with faster members and a faster readout: 43% less time per call at about half the energy.
| Run | GPU | MNIST error | Margin, labels | Hold-out | Ranked ms | mJ above idle | Record |
|---|---|---|---|---|---|---|---|
| 1 | A100 80GB PCIe | 1.809% | +100 | Fashion 88.64% | 1,916.868 | 333,940 | json |
| 2 | A100-SXM4-80GB | 1.810% | +99 | Fashion 88.78% | 1,894.813 | 339,773 | json |
| 3 | A100-SXM4-80GB | 1.821% | +87 | KMNIST 97.22% | 1,898.373 | 321,951 | json |
The three runs used three distinct GPU UUIDs. The worst single MNIST draw was 2.15% (the cap is band + 1.5 = 3.40%), and every call was within 1.5% of its dataset’s median time. verification.json recomputes every verdict, the ranked times and the per-call gates from the raw per-call data in the records.
Source: submission.py, 13,363 bytes, SHA-256 9adc570a02279b34d0766344ec72f5589ff10fae06e1b6b341b58dc351a6445a
(identical to the source_sha256 in all three records). mnist.check_source passes and mnist.review_flags
reports nothing. The file is source only: scalar hyperparameters, no arrays, no fitted data.
x W^T, then RMSNorm with no
learnable gain, then ReLU, then dropout 0.1; hidden layers have no bias. First layer initialised from normalised
training rows. 2,200 updates of 2,048 labelled rows (+1,024 pseudo-labelled rows once cross-teaching starts),
cosine schedule, EMA of the weights.Checked against the rules in mnist.py’s docstring:
classify runs _load (copies the new draw and recomputes its tangents)
and _reset (weights, EMA, Muon momentum, AdamW state, update counter, pseudo-labels, pools, init generator seed)
before any update. The warm-up call only compiles and captures graphs; nothing learned on it reaches the timed
call, and every call runs in a fresh process anyway.train_x, train_y, test_x. Query rows are used without labels (pseudo-labels, kNN graph), as
in the D3 record. Query labels never enter the method.mnist.draw), scored outside the method; nothing derived from those draws is in the file. The
per-layer bias and gain tensors are allocated but unused (they stay at their reset values 0 and 1)./tmp (set at call time), which the scorer clears between calls.dev/quick.py, seeds 401-416, not official scores)| Step | Mean error | ms per call |
|---|---|---|
| GELU MLPs with gain and bias, 2,500 updates (current D5 record, PR #118) | 1.803% | 3,349 |
| + in-graph hash dropout, fused tangent jitter, block readout | 1.787-1.804% | 2,950-2,977 |
| + ReLU, no hidden gain or bias | 1.842% | 2,187 |
| + 2,200 updates (this file) | 1.838-1.849% | 1,893-1,949 |
Removing the gain and bias takes the norm block from 143 to 47 us per layer (their gradients are column sums over all rows, which Inductor cannot fuse with the row-wise norm backward). With these members, a re-tune of tangent strength, Muon learning rate, weight decay, cross-teaching coverage and cadence, dropout and member count found nothing beyond noise.
From mnist-a100:
python run_modal.py submissions/tangent-relu-d5-20261012/submission.py:classify --difficulty 5 --runs 3 --json out/d5-tangent-relu