[megatron] support mlp_padding_free & sp; refactor TransformerLayer#62

Jintao-Huang merged 3 commits intomodelscope:mainmodelscope/mcore-bridge:mainfrom

Jintao-Huang:mlp_padding_free_support_spJintao-Huang/mcore-bridge:mlp_padding_free_support_spCopy head branch name to clipboard