MTP Training Support

[zhoutong-hai]

Hi I wonder if MTP training is already supported? I noticed that Megatron-LM has it and want to check if there is additional config for MTP? Thanks NVIDIA/Megatron-LM@dc385c7#diff-8f7acbd2608d54e2faf8653c0d144c718cd78bcb6a53430c35e81199c6c6651a

[hjh0119]

#6496