Skip to content

Pull requests: linkedin/Liger-Kernel

New pull request New
60 Open 879 Closed
60 Open 879 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

[CuteDSL] Add CuTe DSL cross-entropy kernel and CuteDSL integration scaffolding (B200)
#1279 opened Jul 1, 2026 by justinhh4 Contributor Loading…
Add CuTe DSL implementation of the swiglu liger kernel
#1277 opened Jun 29, 2026 by Celaena24 Loading…
3 tasks done
1
Add tiled MLP patching via the transformers register_patch_mapping API
#1276 opened Jun 28, 2026 by kashif Contributor Loading…
1
Add infer_device_arch() for coarse GPU architecture detection
#1273 opened Jun 27, 2026 by yueyiming2009 Collaborator Loading…
2 of 3 tasks
8
perf(swiglu): architecture-aware column tiling for Blackwell (B200)
#1271 opened Jun 26, 2026 by Celaena24 Loading…
3 tasks
6
perf(flce): fuse grad_weight accumulation via addmm_ (cuBLAS beta=1) + memory cleanups
#1270 opened Jun 26, 2026 by justinhh4 Contributor Loading…
1
[cuTile] Add rope/qwen2vl_mrope/kl_div/group_norm/multi_token_attention
#1269 opened Jun 26, 2026 by xjmxyt Contributor Loading…
1 of 3 tasks
perf(ce): single-shot dx_y gradient correction
#1268 opened Jun 25, 2026 by justinhh4 Contributor Loading…
perf(ce): dtype-aware num_warps (Blackwell-gated)
#1267 opened Jun 25, 2026 by justinhh4 Contributor Loading…
6
[NPU] refact fused_neighborhood_attention
#1265 opened Jun 22, 2026 by sunyi0505 Contributor Loading…
3 tasks done
11
Add DeepSeek 3.2 support
#1264 opened Jun 20, 2026 by michaelroyzen Contributor Loading…
feat: optimize fused linear JSD memory and runtime
#1261 opened Jun 14, 2026 by d7chong Loading…
3 tasks done
5
Cast GroupNorm fp16 backward gradient accumulators to fp16 (not bfloat16)
#1253 opened Jun 4, 2026 by lollinng Loading…
Fix softmax multi-block path for n_cols > 65536 (unreachable + non-compiling)
#1252 opened Jun 4, 2026 by lollinng Loading…
Add inplace=False to liger_cross_entropy to fix upstream gradient corruption (#272)
#1251 opened Jun 4, 2026 by lollinng Loading…
1
perf: optimize grad_weight accumulation with addmm
#1239 opened May 26, 2026 by maskyuanzh Loading…
1 of 3 tasks
1
16
perf: extend RMSNorm block-row dispatch to hidden<=512, lower row threshold
#1238 opened May 25, 2026 by alifurkanstahl Loading…
2 of 3 tasks
fix LigerORPOTrainer bug in none fsdp setup
#1230 opened May 19, 2026 by Ola-Vish • Draft
1 task done
fix(glm4v): version-aware hidden_size/vocab_size for Glm4vConfig
#1214 opened May 5, 2026 by 1fanwang Member Loading…
1
Reduction
#1182 opened Apr 2, 2026 by eljandoubi Contributor Loading…
[Benchmark]: Add model_config sweep mode and model registry
#1180 opened Apr 1, 2026 by noemotiovon Contributor • Draft
3 tasks
30
Enable SwiGLU patching for Qwen3-VL
#1175 opened Mar 31, 2026 by dongchany Loading…
6
Add FSDP support to TiledMLP by preventing premature resharding during the tiled backward recompute loop.
#1128 opened Mar 4, 2026 by alektebel • Draft
2 tasks done
30
Prevent FLCE backward when reduction="none"
#1126 opened Mar 3, 2026 by nick-knight • Draft
1 of 3 tasks
feat(tiled_mlp): add DDP support via ddp_safe flag
#1125 opened Mar 3, 2026 by Rakshitha-Ireddi Loading…
ProTip! Type g i on any issue or pull request to go back to the issue listing page.