Train MOE Model SFT using megatron as backend

[feirz-tech]

When I train MOE Model SFT using megatron as backend, can I use data packing? If I can, could you tell me how i set the config

[wuxibin89]

Yes, just set model.use_remove_padding=True
https://github.com/verl-project/verl/blob/main/examples/sft/vlm/run_qwen3_vl_2b.sh#L89