How much of flash-attention did AI write?
AI signed 1.7% of the 1,613 commits of flash-attention. Signed by Claude (19), Cursor (6), and GitHub Copilot (3).
Since its first commit, on May 20, 2022. Read on November 16, 2026, at commit 328dd7b. Read it as markdown
Who signed the commits
- People1,586 commits
Commits per month
Each month's commits up to its last one, with those an AI signed. Merges don't count.
- Signed by AI
- People
The figures, month by month
| Month | Commits | Signed by AI | Share |
|---|---|---|---|
| Nov 2026 | 36 | 1 | 2.8% |
| Oct 2026 | 26 | 0 | 0% |
| Sep 2026 | 39 | 11 | 28.2% |
| Aug 2026 | 15 | 1 | 6.7% |
| Jul 2026 | 15 | 1 | 6.7% |
| Jun 2026 | 29 | 1 | 3.4% |
| May 2026 | 27 | 2 | 7.4% |
| Apr 2026 | 40 | 4 | 10% |
| Mar 2026 | 122 | 6 | 4.9% |
| Feb 2026 | 71 | 1 | 1.4% |
| Jan 2026 | 45 | 0 | 0% |
| Dec 2025 | 21 | 0 | 0% |
| Nov 2025 | 38 | 0 | 0% |
| Oct 2025 | 102 | 0 | 0% |
| Sep 2025 | 18 | 0 | 0% |
| Aug 2025 | 38 | 0 | 0% |
| Jul 2025 | 42 | 0 | 0% |
| Jun 2025 | 39 | 0 | 0% |
| May 2025 | 4 | 0 | 0% |
| Apr 2025 | 36 | 0 | 0% |
| Mar 2025 | 37 | 0 | 0% |
| Feb 2025 | 43 | 0 | 0% |
| Jan 2025 | 38 | 0 | 0% |
| Dec 2024 | 16 | 0 | 0% |
The last commits an AI signed
- 34a06b9[FA2] Don't dispatch Is_even_MN kernels when leftpad_k / seqused_k is set (#3075) (#3092)Claude· Nov 5, 2026
- 2a92225[Cute, Sm100] 1CTA MLA forward (dense + sparse top-k), MLA dispatch heuristic, sparse MLA backward at any head count (#2938)Claude· Sep 30, 2026
- 2865a89[CuTe, SM100] Sparse MLA bwd preprocess: stop the lse_log2 store at the sequence's rowsClaude· Sep 23, 2026
- 6a3ecb9[CuTe, SM100] Sparse MLA recompute-P: allow 64 Q headsClaude· Sep 23, 2026
- 2cb0a8e[CuTe, SM100] Sparse MLA O residual: guard padded Q-head rows in the residual storeClaude· Sep 23, 2026
- 71dbd93[CuTe, SM100] Sparse MLA training: dpsum from an O residualClaude· Sep 23, 2026
- 784aee1[CuTe, SM100] Sparse MLA exact max: trim interface.py commentsClaude· Sep 23, 2026
- 93c9aca[CuTe, SM100] Sparse MLA training fwd: exact running softmax maxClaude· Sep 23, 2026
- ce1d831[CuTe, SM100] Sparse MLA bwd: keep P normalization and dS product in fp32Claude· Sep 23, 2026
- 294cfb3[CuTe, SM100] Sparse MLA bwd: in-kernel recompute-P + token-chunked backward (#2816)Claude· Sep 16, 2026
How we know
Every AI coding tool leaves its signature in the commits it makes: a Co-Authored-By line, its own author name, or the merge of its branch. SeeRepo reads those signatures in the repo's history; a commit without one counts as made by people, and merges don't count.
Every AI tool SeeRepo recognizes →Frequently asked questions
Which AI tool signed the most commits in flash-attention?
The proof: its last signed commit, 34a06b9
Since when does AI sign commits in flash-attention?
How many commits does flash-attention have?
See inside your code, today.
Paste your repo and see it drawn, part by part: a medium one in under a minute.
Free, no account.Private repo? Sign in with GitHub
Is it your repo? Add its badge, keep it up to date or take it down →

