Skip to content

[Experimental] MiniMax-H3 LoRA training on B200: 2.5x faster step from three changes, with takeaways for the repo's acceleration - #1680

Draft
TarzanZhao wants to merge 3 commits into
modelscope:mainfrom
TarzanZhao:perf/minimax-h3-lora-training
Draft

[Experimental] MiniMax-H3 LoRA training on B200: 2.5x faster step from three changes, with takeaways for the repo's acceleration#1680
TarzanZhao wants to merge 3 commits into
modelscope:mainfrom
TarzanZhao:perf/minimax-h3-lora-training

Commits

Commits on Sep 9, 2026