Create param groups to handle expert + data case (e.g. param.group = moe_group)

[todo]

bagua/bagua/torch_api/model_parallel/moe/experts.py

Lines 27 to 32 in ac6018d

	# TODO: Create param groups to handle expert + data case (e.g. param.group = moe_group)
	for name, param in expert.named_parameters():
	param.allreduce = False
	def forward(self, inputs):
	chunks = inputs.chunk(self.num_local_experts, dim=1)

---

This issue was generated by todo based on a TODO comment in ac6018d. It's been assigned to @liuhatry because they committed the code.