linkedin/Liger-Kernel

View on GitHub

Efficient Triton Kernels for LLM Training

Stars: 6.6k
Forks: 603
Open beginner issues: 0
Indexed issues: 110
Avg merge: 1d 20h
Merged PRs (30d): 47

Dominant language: Python
License: BSD-2-Clause
Last GitHub push: Sep 18, 2026
Latest indexed: Sep 19, 2026
Contributing guide: Contributing guide
Code of conduct: No code of conduct
Beginner labels: help wanted good first issue

Beginner friendly Everything

Open All

110 open issues indexed

Tensor Parallel support Open

high-priority

linkedin/Liger-Kernel#826 · 1 comment · 1 assignee · Opened Jul 20, 2025
Add support for the LFM2 model Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

linkedin/Liger-Kernel#819 · Opened Jul 18, 2025
what is the difference between Triton code written for Liger kernels and generated by torch.compile Open

Difficulty 3/5 1-2 days Newbie friendliness 25/100

linkedin/Liger-Kernel#814 · Opened Jul 17, 2025
Problem when running GRPO with liger kernel. Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

linkedin/Liger-Kernel#810 · 3 comments · 2 reactions · Opened Jul 15, 2025
Why do not apply liger kernal to mlp in the ViT of Qwen2.5-VL? Open

linkedin/Liger-Kernel#805 · 4 comments · 1 assignee · Opened Jul 12, 2025
Layernorm backward regression in triton 3.4.0 release candidate Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

linkedin/Liger-Kernel#801 · 1 comment · 1 reaction · Opened Jul 10, 2025
make test-convergence get Number of mismatched elements Open

Difficulty 2/5 1-3 hours Newbie friendliness 35/100

linkedin/Liger-Kernel#777 · 7 comments · Opened Jun 27, 2025
ZeroDivisionError when finetuning with DeepSpeed + LigerKernel Open

Difficulty 4/5 3-5 days Newbie friendliness 38/100

linkedin/Liger-Kernel#767 · 2 comments · Opened Jun 20, 2025
Need to investigate Gemma3 implementation with Liger Open

Difficulty 4/5 3-5 days Newbie friendliness 30/100

linkedin/Liger-Kernel#757 · Opened Jun 11, 2025
Automating benchmark testing with every pr merged. Open

linkedin/Liger-Kernel#746 · 4 reactions · 1 assignee · Opened Jun 5, 2025
[RFC] Logits numerical issues in convergence test Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

linkedin/Liger-Kernel#742 · 5 reactions · Opened Jun 2, 2025
Grouped Latent Attention Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

linkedin/Liger-Kernel#734 · 2 comments · 3 reactions · Opened May 28, 2025
CI failure tracker Open

Difficulty 4/5 3-5 days Newbie friendliness 20/100

linkedin/Liger-Kernel#729 · 4 comments · Opened May 26, 2025
I have triton 3.3.0 but it shows this error : liger-kernel 0.5.8 depends on triton>=2.3.1 Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

linkedin/Liger-Kernel#682 · 2 comments · 6 reactions · Opened Apr 24, 2025
Support for windows build Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

linkedin/Liger-Kernel#670 · 2 reactions · Opened Apr 16, 2025
No Significant Improvement Observed in Model Training Speed Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

linkedin/Liger-Kernel#664 · 1 comment · Opened Apr 11, 2025
DeepSeek Native Sparse Attention (NSA) Kernel Open

feature fun help wanted

Difficulty 5/5 Over a week Newbie friendliness 35/100

linkedin/Liger-Kernel#656 · 8 comments · 4 reactions · Opened Apr 5, 2025
[RFC] More robust revert functions for convergence tests Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

linkedin/Liger-Kernel#636 · 1 reaction · Opened Mar 30, 2025
Generalized PPO loss (& improve current GRPO loss) Open

linkedin/Liger-Kernel#626 · 4 comments · 1 assignee · Opened Mar 24, 2025
[ROCm]: State of Liger Kernel CI on AMD on ROCm 6.3 Open

Difficulty 5/5 Over a week Newbie friendliness 30/100

linkedin/Liger-Kernel#624 · 1 reaction · Opened Mar 22, 2025
[transformers] support DeepSeek V3 Open

enhancement feature good first issue help wanted huggingface

Difficulty 4/5 3-5 days Newbie friendliness 42/100

linkedin/Liger-Kernel#623 · 4 comments · Opened Mar 22, 2025
Support DAPO Chunked loss Open

feature

linkedin/Liger-Kernel#620 · 3 comments · 1 assignee · Opened Mar 21, 2025
GKD trainer + chunked JSD loss + FSDP Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

linkedin/Liger-Kernel#615 · 2 comments · Opened Mar 17, 2025
When enabling naive model parallelism using ```device_map```, the liger-kernel does not work. Open

linkedin/Liger-Kernel#593 · 5 comments · 3 reactions · 1 assignee · Opened Mar 3, 2025
`revert_liger_kernel_to_xxx` can't revert LigerCrossEntropyLoss for transformers>=4.46.1 Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 38/100

linkedin/Liger-Kernel#542 · 7 comments · 2 reactions · Opened Jan 27, 2025
Issue while building from source on ROCM Open

Difficulty 3/5 1-2 days Newbie friendliness 35/100

linkedin/Liger-Kernel#538 · 4 comments · Opened Jan 22, 2025
Support the new Solar architecture Open

enhancement feature huggingface

Difficulty 5/5 Over a week Newbie friendliness 25/100

linkedin/Liger-Kernel#537 · 2 comments · 2 reactions · Opened Jan 22, 2025
Gradient checkpointing for `grad_weight` in LFCE Open

Difficulty 5/5 Over a week Newbie friendliness 30/100

linkedin/Liger-Kernel#533 · 4 comments · 3 reactions · Opened Jan 20, 2025
Megatron Support Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

linkedin/Liger-Kernel#513 · 4 comments · 2 reactions · Opened Jan 4, 2025
`LigerFusedLinearCrossEntropyLoss` Causes Training Loss to Diverge After Reaching ~8 Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

linkedin/Liger-Kernel#512 · 8 comments · 8 reactions · Opened Jan 4, 2025
Consider support liger kernel for internlm model Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

linkedin/Liger-Kernel#505 · Opened Dec 29, 2024
Extending Liger-Kernel Optimizations to Encoder Models Like BER Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

linkedin/Liger-Kernel#500 · 3 reactions · Opened Dec 26, 2024
Potential Optimization for Preference Training with Prefix Sharing Open

linkedin/Liger-Kernel#476 · 2 reactions · 1 assignee · Opened Dec 13, 2024
Support offline `logits` for teacher model Open

linkedin/Liger-Kernel#441 · 5 comments · 1 reaction · 1 assignee · Opened Dec 9, 2024
PreferenceBase with Softcapping? Open

Difficulty 4/5 3-5 days Newbie friendliness 30/100

linkedin/Liger-Kernel#426 · 2 comments · Opened Dec 4, 2024
The accuracy is misaligned when using bf16 Open

Difficulty 4/5 3-5 days Newbie friendliness 28/100

linkedin/Liger-Kernel#414 · 1 comment · 3 reactions · Opened Nov 28, 2024
No Significant Improvement Observed in Model Training Speed Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

linkedin/Liger-Kernel#409 · 8 comments · Opened Nov 27, 2024
Weighted Cross Entropy Loss Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

linkedin/Liger-Kernel#404 · 4 comments · Opened Nov 21, 2024
ValueError: Pointer argument (at 0) cannot be accessed from Triton (cpu tensor?) Open

Difficulty 4/5 3-5 days Newbie friendliness 38/100

linkedin/Liger-Kernel#401 · 1 comment · 1 reaction · Opened Nov 20, 2024
Apple's cross entropy computation Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

linkedin/Liger-Kernel#391 · 8 comments · 5 reactions · Opened Nov 17, 2024
[RFC] Liger FlexChunkLoss: Alignment and Distillation loss Open

linkedin/Liger-Kernel#371 · 25 comments · 10 reactions · 1 assignee · Opened Nov 8, 2024
cross entropy kernel: inplace modification of input Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

linkedin/Liger-Kernel#343 · 2 comments · Opened Nov 4, 2024
Possible support for weighted average loss calculation in FusedLinearCrossEntropy kernel Open

Difficulty 5/5 Over a week Newbie friendliness 42/100

linkedin/Liger-Kernel#338 · 1 comment · Opened Nov 1, 2024
mllama patch modifies nn.LayerNorm globally Open

Difficulty 3/5 1-2 days Newbie friendliness 55/100

linkedin/Liger-Kernel#315 · 3 comments · Opened Oct 19, 2024
2024 Q4 Roadmap Open

Difficulty 2/5 1-3 hours Newbie friendliness 20/100

linkedin/Liger-Kernel#285 · 2 reactions · Opened Sep 30, 2024
In-place operations in triton kernel might result in incorrect gradient calculations Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 35/100

linkedin/Liger-Kernel#272 · 3 comments · 2 reactions · Opened Sep 26, 2024
Loss does not drop when using Liger Kernel at Qwen2.5 Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

linkedin/Liger-Kernel#257 · 14 comments · Opened Sep 19, 2024
Support for Cohere models Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

linkedin/Liger-Kernel#248 · 1 comment · Opened Sep 15, 2024
MoE kernel Open

feature

Difficulty 5/5 Over a week Newbie friendliness 25/100

linkedin/Liger-Kernel#206 · 6 comments · 1 reaction · Opened Sep 4, 2024
torch.compile() throws exception when LigerKernel is used Open

Difficulty 4/5 3-5 days Newbie friendliness 28/100

linkedin/Liger-Kernel#174 · 24 comments · 4 reactions · Opened Aug 29, 2024

linkedin/Liger-Kernel

Get new issues in your inbox