fix(models): use physical boundaries for Qwen packed attention - #5630

fix(models): use physical boundaries for Qwen packed attention#5630

yaoyu-33 wants to merge 4 commits into

mainNVIDIA-NeMo/Megatron-Bridge:mainfrom

yuya/qwen3-vl-packed-attention-tail-paddingNVIDIA-NeMo/Megatron-Bridge:yuya/qwen3-vl-packed-attention-tail-paddingCopy head branch name to clipboard