/Puzzles
About
Team
Investments
Research
Research Index
Build
IncubationsOpen Source
Writing
Paradigm Puzzles
HackathonAPI

Terms, Disclosures, Privacy

LinkedIn, Twitter, Contact

LeaderboardSubmitGitHubAboutAPI
Dan RobinsonParadigm
LeaderboardSubmitGitHubAboutAPI
Dan Robinson

Write the Fastest Attention Kernel

Build the fastest numerically faithful block-sparse attention backend for an NVIDIA H100. Exact causal block-sparse attention, checked against the FP32 reference.

Your kernel implements block_sparse_attn_fwd over CSR sparse descriptors spanning sliding-window, sink, and retrieval workloads. Entries are scored by the geometric mean of family median latencies on official H100 runs. Check out the GitHub repo to explore the harness.

Read MoreSubmit Your Strategy
keysqueries

Submissions are closed. This challenge has ended. Final standings below — the top 20 finalists were rescored on a fresh hidden seed, median of three official H100 runs.

Leaderboard

Ranked by geometric mean of family median latencies (lower is better)

Show
RankAuthorStrategyGeo MeanWorst FamilyAttempts
#1@0xhellnoNew Dawn0.47 ms0.49 ms8
#2@realbarnakissson of anton 50.50 ms0.50 ms4
#3@jinmo123v123-split-causal-v40.60 ms0.61 ms10
#4@evmknowskek_attn0.69 ms0.72 ms4
#5@sissississi_013usagi?0.74 ms0.78 ms8
#6@that1nerdymax verstappen0.82 ms0.87 ms15
#7@zhimao_liuAskSurf0.87 ms0.93 ms7
#8@0xospreyPowered by AgentCash.dev0.91 ms0.98 ms8
#9@jcmoheddotz0.92 ms0.99 ms8
#10@FrostForgerTriton Press II0.98 ms1.00 ms5
Page 1 of 5