DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling
Cuda · active 2025-04-06 → 2026-08-08 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 490 days from 2025-04-06 to 2026-08-08. Pushes: 110 total, peak 13 in a day. Pull requests: 83 total, peak 12 in a day. Issues: 81 total, peak 5 in a day. Comments: 269 total, peak 10 in a day. Stars: 713 total, peak 40 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 |
|---|---|---|---|---|
| LyricZhao | 173 | 77 | 18 | 69 |
| fzyzcjy | 32 | 0 | 0 | 18 |
| RayWang96 | 30 | 9 | 10 | 7 |
| zheanxu | 23 | 8 | 3 | 8 |
| yewentao256 | 17 | 0 | 0 | 13 |
| yurekami | 16 | 0 | 15 | 1 |
| ko3n1g | 13 | 9 | 2 | 2 |
| qinqinwo | 10 | 0 | 0 | 10 |
| LJC00118 | 10 | 5 | 2 | 3 |
| Sulfur6 | 10 | 0 | 1 | 5 |
| Wangzheee | 9 | 0 | 2 | 6 |
| lucifer1004 | 9 | 0 | 0 | 3 |
| soundOfDestiny | 7 | 0 | 0 | 3 |
| yukuai26 | 6 | 0 | 2 | 4 |
| FlamingoPg | 6 | 0 | 2 | 3 |
| zhonghui-J | 5 | 0 | 1 | 4 |
| chenyu-jiang | 5 | 0 | 0 | 3 |
| zhyncs | 5 | 0 | 0 | 1 |
| simple86 | 5 | 0 | 0 | 2 |
| Chtholly-Boss | 5 | 0 | 0 | 4 |
Recent activity
Latest issues, pull requests and releases
- Pull request#353Barry-Delaney2026-06-04 05:56
- Issue comment#323Stone7499902262026-05-19 03:12feat: MegaMOE adaptation for SM90
- Issue comment#323MikeFang-dev2026-05-15 03:48feat: MegaMOE adaptation for SM90
- Issue comment#324linjiapro2026-05-10 07:53[WIP] feat: Add sm120 support for DeepGEMM
- Issue#330benjibc2026-05-09 20:07[SM100/B300][BF16] bf16_gemm_nt failure with prebuilt wheel, fixed by rebuilding against current PyTorch
- Pull request#327kuishou682026-05-05 02:08
- Issue comment#318linjiapro2026-05-02 05:38Support SM120
- Issue comment#318leavelet2026-05-01 08:46Support SM120
- Issue comment#236scr00ge-002026-04-29 15:06Feature Request: Support sm_120 ( 5090 and blackwell 6000 pro )
- Issue comment#313zheanxu2026-04-27 05:45[Question] Is there a simpler implementation for the UTCCP transposer?
- Issue comment#318jasl2026-04-26 15:48Support SM120
- Issue comment#318wuwenthink2026-04-25 06:59[PoC] Support SM120
- Pull request#316zheanxu2026-04-24 07:34
- Issue#309kuishou682026-04-19 01:05fix: operator precedence bug in pack_ue8m0_to_int assertion (mantissa check always passes)
- Pull request#308Bortlesboat2026-04-18 18:04
- Issue comment#222hpz43112026-04-13 06:56[QUESTION] Can fp8_paged_mqa_logits support next_n > 2?
- Issue comment#301Gregory-Pereira2026-04-11 16:23[Bug] JIT cache lacks cross-process synchronization, can cause failures under multi-process parallelism
- Issue comment#288ajwise92026-03-22 09:04fix: No module named 'scripts.generate_pyi'
- Pull request#292RayWang962026-03-22 08:02
- Issue#221MARD1NO2026-03-09 03:39Can fp8_paged_mqa_logits support non-contiguous layout?
- Issue#291songna20262026-03-07 09:31[question] sm100_fp8_gemm_1d1d_impl: UTCCP transposer for scale factor is strange
- Issue comment#238cailun012026-02-12 07:35fix: k_grouped_fp8_gemm_nt_contiguous crashes with n = 768 on H100
- Pull request#287foreverrookie2026-02-09 14:17
- Issue comment#283cailun012026-01-30 04:03weight gradient kernels for dense and MoE backward 没有grad的反向算子
- Issue#239MARD1NO2026-01-19 03:44[Feature Request] Maybe we can add env var to control whether use min sms?
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 713 stars here means stars gained during the window, not the repo's star count.