DFlash speculative decoding for MiniMax-M2.7 (FSDP2): auto mask-token, FSDP2 resume fixes, per-checkpoint draft export#1621

yeyu-nvidia merged 32 commits into

mainNVIDIA/Model-Optimizer:mainfrom

yeyu/dflash-auto-mask-tokenNVIDIA/Model-Optimizer:yeyu/dflash-auto-mask-tokenCopy head branch name to clipboard