用于 LLM 培训的高效 Triton 内核
Python · active 2024-08-22 → 2026-08-13 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 722 days from 2024-08-22 to 2026-08-13. Pushes: 981 total, peak 21 in a day. Pull requests: 784 total, peak 17 in a day. Issues: 315 total, peak 13 in a day. Comments: 1,567 total, peak 37 in a day. Stars: 3,652 total, peak 656 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 |
|---|---|---|---|---|
| Tcc0403 | 773 | 155 | 55 | 331 |
| ByronHsu | 643 | 151 | 128 | 168 |
| lancerts | 319 | 96 | 67 | 55 |
| shivam15s | 256 | 111 | 58 | 33 |
| vaibhavjindal | 217 | 95 | 43 | 32 |
| shimizust | 192 | 45 | 44 | 40 |
| yundai424 | 149 | 38 | 23 | 36 |
| austin362667 | 133 | 53 | 23 | 31 |
| tyler-romero | 112 | 2 | 15 | 62 |
| hebiao064 | 97 | 37 | 13 | 25 |
| qingquansong | 71 | 9 | 5 | 20 |
| TianHao324 | 66 | 0 | 14 | 34 |
| github-merge-queue[bot] | 63 | 63 | 0 | 0 |
| noemotiovon | 56 | 0 | 9 | 33 |
| github-actions[bot] | 49 | 47 | 1 | 0 |
| jp1924 | 47 | 0 | 3 | 31 |
| AndreSlavescu | 44 | 0 | 10 | 23 |
| tjtanaa | 40 | 13 | 4 | 15 |
| Manan17 | 37 | 0 | 21 | 10 |
| Mecoli1219 | 33 | 2 | 7 | 13 |
Recent activity
Latest issues, pull requests and releases
- Pull request#1349Charlesrizzly2026-08-07 17:24
- Pull request#1339justinhh42026-08-07 16:24
- Pull request#1280Charlesrizzly2026-07-01 04:36
- Issue comment#1249sunyi05052026-06-02 07:02[NPU] remove ineffective params for ASCEND NPU
- Issue comment#1248Mecoli12192026-06-02 00:10fix: add LIGER_FUSED_MOE_AUTOTUNE env to disable fused MoE autotune
- Issue#1246michaelroyzen2026-06-01 02:19MoE improvements in #1179 break DeepSpeed support on B300 due to auto-enabled autotune that can't be disabled
- Issue comment#1235sunyi05052026-05-25 08:07[NPU A3] support fused_moe for A3
- Pull request#1233PKUWZP2026-05-24 01:13
- Pull request#1231sunyi05052026-05-23 11:42
- Issue comment#1194yueyiming20092026-05-22 18:10fix: expose alpha param through LigerFusedLinearDPO public API
- Issue#1200lowdy12026-05-21 06:14[Benchmark] Benchmark probe issue, memory probe and estimation may be inaccurate for some kernels
- Issue comment#1228vaibhavjindal2026-05-18 23:53Add cutile jsd
- Issue comment#1216UserChen6662026-05-16 08:33[Benchmark] add benchmark for grpo_loss kernel
- Issue comment#1224yukiu002026-05-13 11:54Add fused Modulated RMSNorm for DiT-style scale/shift conditioning
- Issue comment#1082yukiu002026-05-02 18:21LigerFusedLinearGRPOLoss produces ~100x larger grad_norm than TRL's non-Liger-Kernel path due to missing vLLM IS correction and other differences
- Pull request#1211sunyi05052026-04-30 15:09
- Issue comment#1211sunyi05052026-04-30 08:27[NPU] Modify the grid calculation logic to support larger batch sizes for group norm.
- Issue comment#1208shivam21992026-04-29 07:11perf(dyt): enable triton.autotune for DyT forward and backward kernels
- Pull request#1208shivam21992026-04-29 05:25
- Pull request#1202kashif2026-04-26 07:01
- Issue comment#1201shivam21992026-04-25 11:32Add Falcon H1 SwiGLU multipliers support (#936)
- Issue comment#1201shivam21992026-04-25 11:03Add Falcon H1 SwiGLU multipliers support (#936)
- Pull request#1201shivam21992026-04-25 10:58
- Issue comment#1186Mecoli12192026-04-24 23:10Gemma4 text model support
- Issue comment#1196Mecoli12192026-04-24 22:27[Gemma 4] Add apply_liger_kernel_to_gemma4_text (dense text, 31B-targeted)
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 3,652 stars here means stars gained during the window, not the repo's star count.