🚀 Efficient implementations of state-of-the-art linear attention models in Torch and Triton - BitNet patch
Python · active 2025-04-05 → 2026-08-07 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 490 days from 2025-04-05 to 2026-08-07. Pushes: 1,019 total, peak 38 in a day. Pull requests: 283 total, peak 16 in a day. Issues: 161 total, peak 7 in a day. Comments: 1,061 total, peak 24 in a day. Stars: 1,169 total, peak 37 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 |
|---|---|---|---|---|
| coderabbitai[bot] | 932 | 0 | 1 | 548 |
| zhiyuan1i | 725 | 504 | 113 | 76 |
| yzhangcs | 551 | 399 | 82 | 41 |
| gemini-code-assist[bot] | 258 | 0 | 0 | 164 |
| sustcsonglin | 192 | 85 | 18 | 38 |
| github-actions[bot] | 66 | 0 | 0 | 40 |
| Copilot | 51 | 0 | 0 | 27 |
| Nathancgy | 9 | 0 | 7 | 1 |
| hypnopump | 8 | 0 | 0 | 4 |
| lucassunalt | 8 | 0 | 0 | 6 |
| mutiann | 6 | 0 | 2 | 2 |
| yangshengaa | 5 | 0 | 0 | 3 |
| maomaocun | 5 | 0 | 0 | 4 |
| AwesomeSeq | 5 | 0 | 5 | 0 |
| Costa-SM | 5 | 0 | 0 | 4 |
| richardodliu | 4 | 0 | 2 | 2 |
| tmct | 4 | 0 | 0 | 2 |
| ReyJerry | 4 | 0 | 0 | 2 |
| 2022tgoel | 4 | 0 | 3 | 1 |
| slowlyC | 4 | 0 | 0 | 3 |
Recent activity
Latest issues, pull requests and releases
- Pull request#1008zhiyuan1i2026-07-07 05:26
- Issue#889github-actions[bot]2026-06-21 02:29[Bug] Non-atomic shared-memory accumulation causes nondeterministic dg gradients in tilelang chunk_bwd
- Pull request#964yzhangcs2026-06-20 22:09
- Pull request#961yzhangcs2026-06-20 20:27
- Issue comment#734alexxchen2026-05-30 06:15[Bug] Multiple FLA kernels crash on H100 (sm_90) with Triton 3.2-3.6
- Issue comment#918Fr0do2026-05-24 14:43Add reference higher-order linear attention
- Issue comment#910Michael-RDev2026-05-21 00:05[Fix] FLA cache window updates and reset behavior
- Issue#906zhiyuan1i2026-05-17 22:46[Bug] `gate_output_correction` skips the tail chunk for `T > 65536`
- Issue#906sablin392026-05-17 15:41[Bug] `gate_output_correction` skips the tail chunk for `T > 65536`
- Issue#903sablin392026-05-17 12:35[Bug] `token_shift` kernel selection logic failed in long packed seqlen with many docs
- Issue comment#900github-actions[bot]2026-05-16 21:41Add Raven model integration
- Pull request#892xylian862026-05-12 19:54
- Pull request#881KevinZeng082026-05-11 18:54
- Pull request#888yzhangcs2026-05-10 21:23
- Pull request#886yzhangcs2026-05-10 13:53
- Issue#106yzhangcs2026-04-26 23:16[RFC] Add YOCO models
- Pull request#861zhiyuan1i2026-04-26 07:06
- Issue comment#860CiprianFlorin-Ifrim2026-04-24 15:56[Fix] GatedDeltaNet step mode crash when conv cache is None on cold Triton autotuner cache
- Issue comment#121younengma2026-04-24 00:52[Bug] AttributeError: 'NoneType' object has no attribute 'start'
- Pull request#853yzhangcs2026-04-21 04:26
- Issue#850yzhangcs2026-04-19 22:13[Bug] Incorrect cumsum dim for naive impl of linear_attn
- Pull request#851zhiyuan1i2026-04-19 11:03
- Issue#850HydrogenSulfate2026-04-19 09:47[Bug] Incorrect cumsum dim for naive impl of linear_attn
- Issue#850HydrogenSulfate2026-04-19 09:47[Bug] Incorrect cumsum dim for naive impl of linear_attn
- Issue comment#790zhiyuan1i2026-04-19 09:19[Bug] Incorrect outputs for chunk_gated_delta_rule_fwd_kernel_h_blockdim64 on blackwell for certain autotune configs
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 1,169 stars here means stars gained during the window, not the repo's star count.