Megatron-DeepSpeed

1cb76a64 - sync the whole Meg-LM fused_kernels sub-tree (#260)

sync the whole Meg-LM fused_kernels sub-tree (#260) * sync the whole Meg-LM fused_kernels sub-tree * author attribution * part 2 Co-authored-by: hyunwoongko <gusdnd852@naver.com>