qwen3.5 moe model merge

[abcdvzz]

Checklist / 检查清单

• I have searched existing issues, and this is a new question or discussion topic. / 我已经搜索过现有的 issues，确认这是一个新的问题与讨论。

Question Description / 问题描述

Hi Team, may I ask how to merge the sft lora checkpoint with the base model, and then use it as new base model for grpo? Or are there other ways to do sft lora -> grpo lora training? Thank you.

[tastelikefeet]

meaning how to merge the training result(lora) with the original model?

[abcdvzz]

yes

[tastelikefeet]

Twinkle does not offer the ability of merging loras, you can use swift to merge lora: https://swift.readthedocs.io/en/latest/Instruction/Export-and-push.html