Bug of MoE checkpoint. #1185

Open

opened

Now the parallelism of attention and moe is decouple. So there are so many bugs in convert megatron checkpoint to hf checkpoint, especially the expert_tensor_parallel. Please arrange for someone to handle it.

to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Assignees

No one assigned

Labels

No labels

No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests