| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| README.md | 2025-10-27 | 4.8 kB | |
| v0.6.3 release source code.tar.gz | 2025-10-27 | 3.8 MB | |
| v0.6.3 release source code.zip | 2025-10-27 | 4.0 MB | |
| Totals: 3 Items | 7.7 MB | 3 | |
Highlights in this release:
New model architecture supports: SmolVLM2, GLM4.5V, InternVL3, Falcon-H1, Qwen-Next
New algorithm: GSPO
What's Changed
- [cross-entropy-loss] Added support for DFT flag by @kashif in https://github.com/linkedin/Liger-Kernel/pull/860
- fix(test): update assertions in GLM4 instance patching tests by @vvvdwbvvv in https://github.com/linkedin/Liger-Kernel/pull/859
- Fix nan loss error for LigerFusedLinearJSDLoss by @ParagEkbote in https://github.com/linkedin/Liger-Kernel/pull/862
- [Cross-entropy] get valid predicted probabilities by @kashif in https://github.com/linkedin/Liger-Kernel/pull/864
- Enhance Docs by @ParagEkbote in https://github.com/linkedin/Liger-Kernel/pull/867
- Add Classifiers for Liger-Kernel by @ParagEkbote in https://github.com/linkedin/Liger-Kernel/pull/869
- docs(mta): suppress invalid sequence syntax warning by @Tcc0403 in https://github.com/linkedin/Liger-Kernel/pull/870
- Add GSPO by @BjarniHaukur in https://github.com/linkedin/Liger-Kernel/pull/845
- Add GLM4.5V support by @vvvdwbvvv in https://github.com/linkedin/Liger-Kernel/pull/863
- A Fix for Issue [#872] by @yshenaw in https://github.com/linkedin/Liger-Kernel/pull/879
- Add pytest coverage for liger-kernel by @ParagEkbote in https://github.com/linkedin/Liger-Kernel/pull/876
- Replace all torch_dtype with dtype by @Tcc0403 in https://github.com/linkedin/Liger-Kernel/pull/881
- Update Dev Dependencies by @ParagEkbote in https://github.com/linkedin/Liger-Kernel/pull/886
- Fixed AMD CI issue [#793] by @DevManpreet5 in https://github.com/linkedin/Liger-Kernel/pull/887
- fix(layernorm): remove
n_colsupcasting for for torch.compile by @Tcc0403 in https://github.com/linkedin/Liger-Kernel/pull/884 - Fix tests and CI by @vaibhavjindal in https://github.com/linkedin/Liger-Kernel/pull/882
- Remove daily test cron job by @vaibhavjindal in https://github.com/linkedin/Liger-Kernel/pull/890
- [UT] [XPU] Modify the test cases of XPU for triton3.5 by @YangKai0616 in https://github.com/linkedin/Liger-Kernel/pull/889
- Add InternVL3 support by @MilkClouds in https://github.com/linkedin/Liger-Kernel/pull/878
- fix(flce): add
shift_labelsas eval mode loss condition by @Tcc0403 in https://github.com/linkedin/Liger-Kernel/pull/888 - Add support of Falcon-H1 models for liger kernels by @puneeshkhanna in https://github.com/linkedin/Liger-Kernel/pull/900
- Don't deploy mkdocs to fix benchmarking by @vaibhavjindal in https://github.com/linkedin/Liger-Kernel/pull/904
- Disable mllama multimodal test in transformers<4.51.0 by @Tcc0403 in https://github.com/linkedin/Liger-Kernel/pull/899
- Add flce forward for FalconH1ForCausalLM and missing tests by @Tcc0403 in https://github.com/linkedin/Liger-Kernel/pull/903
- feat(ce,flce): decouple gradients computation for no_grad mode by @Tcc0403 in https://github.com/linkedin/Liger-Kernel/pull/894
- fix(llama4): Get correct swiglu patch target for llama4 moe layer by @alenawang in https://github.com/linkedin/Liger-Kernel/pull/907
- Add PolyNorm operator by @0xtoward in https://github.com/linkedin/Liger-Kernel/pull/901
- Copy and paste benchmarks before and after gh-pages deployment by @vaibhavjindal in https://github.com/linkedin/Liger-Kernel/pull/909
- Filter out redundant ops/allocations in no_grad mode by @Tcc0403 in https://github.com/linkedin/Liger-Kernel/pull/906
- Add support for Qwen3Next model with Liger kernels by @vvvdwbvvv in https://github.com/linkedin/Liger-Kernel/pull/912
- refactor(convergence-test): remove unnecessary print by @Tcc0403 in https://github.com/linkedin/Liger-Kernel/pull/913
- Enabled the tests glm4v/glm4v_moe for XPU and Fixed the monkey patch error by @YangKai0616 in https://github.com/linkedin/Liger-Kernel/pull/914
- [Test][XPU] Added gpu cache cleaning for XPU devices by @Egor-Krivov in https://github.com/linkedin/Liger-Kernel/pull/917
- Add SmolVLM2 support by @MilkClouds in https://github.com/linkedin/Liger-Kernel/pull/919
New Contributors
- @BjarniHaukur made their first contribution in https://github.com/linkedin/Liger-Kernel/pull/845
- @yshenaw made their first contribution in https://github.com/linkedin/Liger-Kernel/pull/879
- @DevManpreet5 made their first contribution in https://github.com/linkedin/Liger-Kernel/pull/887
- @MilkClouds made their first contribution in https://github.com/linkedin/Liger-Kernel/pull/878
- @puneeshkhanna made their first contribution in https://github.com/linkedin/Liger-Kernel/pull/900
- @alenawang made their first contribution in https://github.com/linkedin/Liger-Kernel/pull/907
- @0xtoward made their first contribution in https://github.com/linkedin/Liger-Kernel/pull/901
Full Changelog: https://github.com/linkedin/Liger-Kernel/compare/v0.6.2...v0.6.3