CUDA kernels for linear attention variants, written in CuTe DSL and CUTLASS C++.
active 2026-04-02 → 2026-08-14 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 135 days from 2026-04-02 to 2026-08-14. Pushes: 71 total, peak 11 in a day. Pull requests: 8 total, peak 2 in a day. Issues: 19 total, peak 7 in a day. Comments: 59 total, peak 15 in a day. Stars: 99 total, peak 41 in a day.
- Pushes
- Pull requests
- Issues
- Comments
- Stars
Stars, PRs, issues and forks are under-captured in the later part of this window. GH Archive progressively stopped capturing non-push events during 2026 — −95% or worse by the end of the window. Every series here except Pushes fades for that reason, so a decline above reflects the archive, not this repository. Pushes stay reliable throughout, so read them, and the contributor counts derived from them, as the real signal. Data health has the measurements.
Top contributors
Pushes, PRs, issues, reviews and comments — stars and forks excluded, so this is contribution rather than popularity
| Contributor | Contributions | Pushes | PRs | Comments |
|---|---|---|---|---|
| KevinZeng08 | 64 | 38 | 2 | 14 |
| icavan | 50 | 12 | 2 | 7 |
| gemini-code-assist[bot] | 23 | 0 | 0 | 17 |
| Copilot | 10 | 3 | 0 | 3 |
| icavanyu | 9 | 9 | 0 | 0 |
| cherhh | 6 | 0 | 0 | 2 |
| yuan-luo | 5 | 0 | 0 | 5 |
| tongke6 | 5 | 5 | 0 | 0 |
| zheyang0825 | 4 | 3 | 1 | 0 |
| yechenzhi | 3 | 0 | 0 | 3 |
| sjmshsh | 2 | 0 | 0 | 2 |
| meinie0826 | 2 | 0 | 1 | 1 |
| jorel666 | 1 | 0 | 0 | 1 |
| Azir9 | 1 | 0 | 1 | 0 |
| piDack | 1 | 0 | 0 | 1 |
| fkuner | 1 | 1 | 0 | 0 |
| Emre-Dinc | 1 | 0 | 0 | 1 |
| venturevd | 1 | 0 | 0 | 1 |
| consigcody94 | 1 | 0 | 0 | 1 |
| higgsboson1710 | 1 | 0 | 1 | 0 |
Recent activity
Latest issues, pull requests and releases
- Issue#92icavan2026-06-14 04:34Rewrite SM100 KDA Forward in CuteDSL
- Issue#81icavan2026-05-25 03:32[Feature] Add GVA support for Lightning
- Pull request#79icavan2026-05-24 01:26
- Issue comment#75Emre-Dinc2026-05-23 11:19fix: wire h0_indices into Lightning Attention decode for state-pool indexing
- Issue comment#65sjmshsh2026-05-23 04:07[KDA] sm100 GVA enhance
- Issue comment#64KevinZeng082026-05-09 02:44[KDA] sm90 GVA enhance
- Issue comment#66KevinZeng082026-05-09 01:25fix internal output_final_state wrapper issue
- Issue comment#66yechenzhi2026-05-08 03:54fix internal output_final_state wrapper issue
- Issue#10icavan2026-05-03 14:51Larger chunk size and 2-CTA on SM10X for improved throughput
- Issue comment#62jorel6662026-05-03 03:37Draft: feat(kda, sm90): add KDA GVA forward support
- Issue comment#55yechenzhi2026-05-03 03:37[KDA] Grouped Value Attention (GVA) support
- Issue#50icavan2026-04-25 14:20WIP-RFC: Multi-GPU Unified Interface for cuLA
- Issue comment#55KevinZeng082026-04-19 16:36[KDA] Grouped Value Attention (GVA) support
- Issue comment#47icavan2026-04-18 16:16KDA_prefill_sm89
- Issue#50icavan2026-04-13 02:56WIP-RFC: Multi-GPU Unified Interface for cuLA
- Issue#49icavan2026-04-11 13:01Explore high-occupancy kernel variants with smaller chunk sizes
- Issue#49icavan2026-04-11 13:01Explore high-occupancy kernel variants with smaller chunk sizes
- Issue#49icavan2026-04-11 13:01Explore high-occupancy kernel variants with smaller chunk sizes
- Issue#49icavan2026-04-11 12:59Explore high-occupancy kernel variants with smaller chunk sizes
- Pull request#47Azir92026-04-11 01:36
- Pull request#45KevinZeng082026-04-10 07:27
- Pull request#44zheyang08252026-04-09 03:54
- Issue comment#41piDack2026-04-09 03:34Hopper fused KDA benchmark fails on H200 when `D=64` but works when `D=128`
- Issue comment#43meinie08262026-04-08 10:10fix(kernel): fix NaN in use_gate_in_kernel=True path (cache_key, copy atom, pipeline race, tail mask)
- Pull request#38meinie08262026-04-08 09:20
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 99 stars here means stars gained during the window, not the repo's star count.