Skip to content

Conversation

@kashif
Copy link
Contributor

@kashif kashif commented Jan 15, 2026

Summary

Some changes for transformers v5 issues.

Testing Done

  • Hardware Type:
  • run make test to ensure correctness
  • run make checkstyle to ensure code style
  • run make test-convergence to ensure convergence

kashif and others added 4 commits January 15, 2026 19:40
- Add liger_hunyuan_v1_moe_experts_forward with fused SiLU*mul optimization
- Add liger_glm4v_moe_experts_forward with fused SiLU*mul optimization
- Update monkey patching to use hasattr for MoE detection instead of isinstance
- Fix test config to enable MoE on all layers (first_k_dense_replace=0)
- Fix test assertions to handle MoE vs dense layers correctly
- Add LigerHunyuanV1SwiGLUMLP import in Hunyuan test
- Remove unused Glm4vMoeTextMoE variables

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <[email protected]>
- Remove HybridCache import in gemma2.py, use string type hint instead
- Add liger_qwen3_moe_experts_forward with fused SiLU*mul optimization
- Update Qwen3 MoE monkey patching to use experts forward
- Update Qwen3 MoE test to check for experts forward patching
- Simplify instance-level patching using hasattr instead of try-except

Cherry-picked and adapted from: 6ab0886

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <[email protected]>
@kashif kashif closed this Jan 15, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant