๐ณ Efficient Triton implementations for "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention"
active 2025-03-01 โ 2026-05-06 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 432 days from 2025-03-01 to 2026-05-06. Pushes: 20 total, peak 9 in a day. Pull requests: 3 total, peak 2 in a day. Issues: 30 total, peak 3 in a day. Comments: 42 total, peak 5 in a day. Stars: 245 total, peak 17 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 โ โ95% 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 โ stars and forks excluded, so this is contribution rather than popularity
| Contributor | Contributions | Pushes | PRs | Comments |
|---|---|---|---|---|
| github-actions[bot] | 37 | 0 | 0 | 24 |
| yzhangcs | 19 | 17 | 0 | 0 |
| gemini-code-assist[bot] | 6 | 0 | 0 | 5 |
| Hanyuezhuohua | 6 | 3 | 2 | 1 |
| imhuim982 | 4 | 0 | 0 | 2 |
| haok1402 | 4 | 0 | 0 | 2 |
| PXHnb666 | 2 | 0 | 0 | 0 |
| longlong-yu | 2 | 0 | 0 | 0 |
| adityavavreNVDA | 2 | 0 | 0 | 0 |
| p81sunshine | 2 | 0 | 0 | 2 |
| coderabbitai[bot] | 1 | 0 | 0 | 1 |
| HarounH | 1 | 0 | 0 | 0 |
| weixuansun | 1 | 0 | 0 | 0 |
| JHL328 | 1 | 0 | 0 | 1 |
| Youhe-Jiang | 1 | 0 | 0 | 0 |
| mingfa-w | 1 | 0 | 1 | 0 |
| Ahmedhasssan-aig | 1 | 0 | 0 | 1 |
| kiscad | 1 | 0 | 0 | 0 |
| lengxia | 1 | 0 | 0 | 1 |
| poryfly | 1 | 0 | 0 | 0 |
Recent activity
Latest issues, pull requests and releases
- Issue#34longlong-yu2025-11-28 14:09[Bug] Incorrect dv address calculation in parallel_nsa_compression_bwd_kernel_dkv
- Issue#34longlong-yu2025-11-28 14:09[Bug] Incorrect dv address calculation in parallel_nsa_compression_bwd_kernel_dkv
- Issue#32github-actions[bot]2025-11-02 01:17[Bug] FAILED tests/test_nsa.py
- Issue comment#32github-actions[bot]2025-11-02 01:17[Bug] FAILED tests/test_nsa.py
- Issue#32github-actions[bot]2025-10-26 01:16[Bug] FAILED tests/test_nsa.py
- Issue#33poryfly2025-10-20 09:13The meaning of the data in benchmark is that T must reach a certain length to generate returns, right?
- Issue#32PXHnb6662025-09-20 08:38[Bug] FAILED tests/test_nsa.py
- Issue comment#27lengxia2025-09-15 02:09[Bug] Not able to match the performance numbers of the NSA kernel
- Issue comment#31haok14022025-08-31 21:49TypeError: parallel_nsa() missing 1 required positional argument: 'g_cmp'
- Issue#31haok14022025-08-31 21:49TypeError: parallel_nsa() missing 1 required positional argument: 'g_cmp'
- Issue comment#29haok14022025-08-31 21:24[RFC] An error occurred when running pytest tests/test_nsa.py in the native-sparse-attention folder
- Issue#31haok14022025-08-31 21:18TypeError: parallel_nsa() missing 1 required positional argument: 'g_cmp'
- Pull request#30mingfa-w2025-08-29 08:37
- Issue#26github-actions[bot]2025-08-17 01:23[Bug] Head Dimension is always set to the default value 64
- Issue comment#27github-actions[bot]2025-08-17 01:23[Bug] Not able to match the performance numbers of the NSA kernel
- Issue#29PXHnb6662025-08-05 14:24[RFC] An error occurred when running pytest tests/test_nsa.py in the native-sparse-attention folder
- Issue comment#25sehunfromdaegu2025-08-04 04:20[Bug] Not able to run the basic test
- Issue comment#26github-actions[bot]2025-08-03 01:38[Bug] Head Dimension is always set to the default value 64
- Issue comment#27github-actions[bot]2025-08-03 01:38[Bug] Not able to match the performance numbers of the NSA kernel
- Issue#28Youhe-Jiang2025-08-02 11:43[RFC] Does NSA support ultra long training length (e.g., 128K)?
- Issue comment#25github-actions[bot]2025-07-27 01:28[Bug] Not able to run the basic test
- Issue#25github-actions[bot]2025-07-27 01:28[Bug] Not able to run the basic test
- Issue#27adityavavreNVDA2025-07-03 21:20[Bug] Not able to match the performance numbers of the NSA kernel
- Issue#26adityavavreNVDA2025-07-01 07:35[Bug] Head Dimension is always set to the default value 64
- Issue comment#25Ahmedhasssan-aig2025-06-16 17:47[Bug] Not able to run the basic test
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals โ 245 stars here means stars gained during the window, not the repo's star count.