|
Has anyone else noticed that the 50 leaderboard MLPs are unusually easy / low-variance?
|
|
1
|
22
|
August 10, 2026
|
|
Request to extend the deadline of Phase 1
|
|
10
|
1062
|
August 9, 2026
|
|
Phase 1 update: flopscope v0.10.0, cost-model fixes, residual-time safeguards, and updated deadlines
|
|
8
|
938
|
August 8, 2026
|
|
[flopscope 0.10.0] ALL of `flopscope.stats` silently promotes float32 to float64, doubling the bill downstream
|
|
1
|
390
|
August 7, 2026
|
|
[flopscope] budget_summary_dict() cost grows with process history — silently inflates residual and can cost the 0.1 multiplier floor
|
|
1
|
538
|
August 7, 2026
|
|
Recommendation: restoring the estimation framing (neutralizing the wall-time compute channel)
|
|
1
|
587
|
August 7, 2026
|
|
Potential flopscope accounting bypass bug
|
|
3
|
901
|
August 7, 2026
|
|
How exactly is phase 1 submission validation going to work?
|
|
1
|
238
|
August 7, 2026
|
|
A bet on a floor, filed before the reveal
|
|
4
|
800
|
August 7, 2026
|
|
Will the private re-evaluation verify instrumented share?
|
|
1
|
357
|
August 5, 2026
|
|
Rules clarification needed on compute budget exceedance
|
|
1
|
317
|
August 5, 2026
|
|
Phase 1 nomination interface and the Aug 17 write-up — where do they live?
|
|
0
|
282
|
August 5, 2026
|
|
Grading queue backlog since the v0.10.0 rollout
|
|
2
|
264
|
August 4, 2026
|
|
Phase 1 write-up: Stabilizing cumulant propagation at depth 32 — a trajectory-calibrated moment chain, with an error budget (submission #314695)
|
|
1
|
1087
|
August 4, 2026
|
|
Rules clarification: are operations on the NumPy-backed arrays reachable through flopscope.numpy inside the intended accounting boundary?
|
|
5
|
615
|
August 3, 2026
|
|
Phase 1 final evaluation question: Overfitting to public instances
|
|
3
|
632
|
August 3, 2026
|
|
Urgent: submission limit is rolling 24h, not the published UTC-day reset
|
|
1
|
309
|
July 31, 2026
|
|
Phase I Submission: Structure-Aware Estimation for Random ReLU MLPs
|
|
0
|
832
|
July 27, 2026
|
|
Smoke wall time limit is different from evaluation limit?
|
|
0
|
300
|
July 26, 2026
|
|
Cost-model question: releasing temporaries of counted ops is charged to residual — and a request to expose out= on fnp.matmul
|
|
0
|
437
|
July 26, 2026
|
|
There is a bug in flopscope.numpy.linalg.solve that undercounts batched right-hand sides
|
|
5
|
830
|
July 25, 2026
|
|
Submissions score 50/50 with 0 failures but flagged "Evaluation could not complete"
|
|
1
|
434
|
July 23, 2026
|
|
[whest run] Residual Wall Time in the budget-breakdown panel reads ~39,807 s under --runner subprocess
|
|
0
|
385
|
July 23, 2026
|
|
Phase 1 write-up: A variance/bias budget for deep white-box mean estimation (submission 317660)
|
|
0
|
1194
|
July 21, 2026
|
|
Flopscope v0.8.0 Release Candidate Available!
|
|
12
|
1943
|
July 20, 2026
|
|
📹 Townhall Summary & Recording
|
|
0
|
1022
|
July 17, 2026
|
|
[flopscope] Bug report: empty contractions receive negative FLOP charges (budget increases)
|
|
1
|
540
|
July 17, 2026
|
|
Live Q&A with ARC Organisers | Join the Townhall on July 6
|
|
3
|
898
|
July 16, 2026
|
|
Leaderboard issue!
|
|
0
|
1567
|
July 12, 2026
|
|
Unbiased randomized-QMC + Rao-Blackwell for post-ReLU activation means — method, unbiasedness proof, and where the frontier is
|
|
1
|
1826
|
July 7, 2026
|