快速且节省内存的精度 关注 Fast and memory-efficient exact attention
Python · active 2025-04-06 → 2026-08-09 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 491 days from 2025-04-06 to 2026-08-09. Pushes: 642 total, peak 43 in a day. Pull requests: 257 total, peak 6 in a day. Issues: 238 total, peak 4 in a day. Comments: 1,290 total, peak 20 in a day. Stars: 2,909 total, peak 26 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 |
|---|---|---|---|---|
| tridao | 544 | 233 | 27 | 215 |
| drisspg | 502 | 242 | 66 | 100 |
| jayhshah | 146 | 88 | 15 | 31 |
| johnnynunez | 69 | 0 | 11 | 54 |
| ko3n1g | 61 | 57 | 4 | 0 |
| Johnsonms | 48 | 18 | 8 | 13 |
| v0i0 | 28 | 2 | 3 | 10 |
| reubenconducts | 28 | 0 | 11 | 12 |
| guilhermeleobas | 21 | 0 | 0 | 18 |
| Copilot | 20 | 0 | 0 | 11 |
| Alkaid-Benetnash | 18 | 0 | 5 | 7 |
| janeyx99 | 17 | 0 | 0 | 9 |
| henrylhtsang | 15 | 0 | 5 | 7 |
| micmelesse | 14 | 0 | 3 | 11 |
| Edenzzzz | 13 | 0 | 1 | 7 |
| chatgpt-codex-connector[bot] | 13 | 0 | 0 | 12 |
| vadimkantorov | 12 | 0 | 0 | 10 |
| Luciennnnnnn | 12 | 0 | 1 | 9 |
| blake-snc | 11 | 0 | 2 | 7 |
| loscrossos | 10 | 0 | 0 | 9 |
Recent activity
Latest issues, pull requests and releases
- Issue comment#2653hebo12212026-07-24 06:03_flash_attn_fwd/_flash_attn_bwd: DSLRuntimeError — window_size_left/right expects Int32 but gets plain Python int
- Issue comment#2427alex-ht2026-07-01 05:09Support head_dim=512 for Gemma 4 global attention layers
- Pull request#2662sryap2026-06-16 22:29
- Issue comment#2635Johnsonms2026-06-15 00:16Training Extremely Slow on Qwen3.5-35B-A3B + 8×B300 (280s/step), py-spy Shows FlashAttention-4 CUTLASS JIT Compilation During Backward
- Issue comment#2578Edenzzzz2026-06-06 05:24[Cute,Fwd,Sm100] Fix FP8 precision loss at tile boundaries
- Pull request#2629jayhshah2026-06-05 23:36
- Issue comment#2416vient2026-06-05 21:19Fix SM120 forward pass crash: parent __init__ overwrites arch, enabling unsupported TMA path
- Pull request#2627Johnsonms2026-06-04 22:51
- Pull request#2626Luosuu2026-06-04 22:16
- Pull request#2620micmelesse2026-06-02 22:46
- Issue comment#2611Johnsonms2026-06-01 02:29[FA4] cute varlen backward: int32 overflow in `_bwd_preprocess` corrupts `dQ/dK/dV` when `dout` has a large outer stride
- Issue comment#2611jayhshah2026-06-01 01:50[FA4] cute varlen backward: int32 overflow in `_bwd_preprocess` corrupts `dQ/dK/dV` when `dout` has a large outer stride
- Issue#2607trufflez5652026-05-31 00:45buy crypto with credit card without OTP verification
- Pull request#2595Luosuu2026-05-28 23:41
- Issue comment#2588chejh-amd2026-05-26 05:19[ROCm] gfx1201 compilation failure - v_cmp_u_f32 invalid
- Issue comment#2588UFO05062026-05-26 02:17[ROCm] gfx1201 compilation failure - v_cmp_u_f32 invalid
- Issue#2591Johnsonms2026-05-26 01:26[FA4, CuTe, SM100] `_flash_attn_fwd` fails at launch for `head_dim < 32` on B200
- Pull request#2594Johnsonms2026-05-26 01:19
- Issue comment#2589lingolin1282026-05-25 08:02Use is_family_of for sm_90 and sm_103 arch checks
- Issue#2588UFO05062026-05-25 05:24[ROCm] gfx1201 compilation failure - v_cmp_u_f32 invalid
- Pull request#2572ocss8842026-05-24 22:55
- Pull request#2261reubenconducts2026-05-24 01:26
- Issue comment#188anasmohiuddinsyed-bit2026-05-24 01:05Error in setup.py when installing the package from a fresh conda environment
- Pull request#2585factnn2026-05-23 12:13
- Pull request#2584sryap2026-05-22 22:40
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 2,909 stars here means stars gained during the window, not the repo's star count.