Checklist / 检查清单
Bug Description / Bug 描述
moe_router_load_balancing_type has inconsistent type declarations across different code locations
How to Reproduce / 如何复现
In megatron_args, list
|
moe_router_load_balancing_type: Optional[List[str]] = None |
In model_config, string
|
moe_router_load_balancing_type: Literal['aux_loss', 'seq_aux_loss', 'global_aux_loss', 'sinkhorn', |
|
'none'] = 'aux_loss' |
Additional Information / 补充信息
No response
Checklist / 检查清单
Bug Description / Bug 描述
moe_router_load_balancing_type has inconsistent type declarations across different code locations
How to Reproduce / 如何复现
In megatron_args, list
ms-swift/swift/megatron/arguments/megatron_args.py
Line 511 in f0aecc0
In model_config, string
ms-swift/swift/megatron/model/model_config.py
Lines 175 to 176 in f0aecc0
Additional Information / 补充信息
No response