perf(MoE): Use TE quant/dequant for SwiGLU fp8 input store to improve performance and stability#1753

xiaoxi-wangfj wants to merge 3 commits intoNVIDIA:mainNVIDIA/Megatron-LM:mainfrom

021ai:optimize-swiglu-input-fp8-quant021ai/Megatron-LM:optimize-swiglu-input-fp8-quantCopy head branch name to clipboard