perf(speculative): speed up DFlash/DSpark drafter training 3.49x - #2279

perf(speculative): speed up DFlash/DSpark drafter training 3.49x#2279

h-guo18 wants to merge 13 commits into

mainNVIDIA/Model-Optimizer:mainfrom

haoguo/dflash-training-perfNVIDIA/Model-Optimizer:haoguo/dflash-training-perfCopy head branch name to clipboard