Skip to content

支持Qwen3 MoE的Megatron LoRA训练 #4126

Closed
@Bond-H

Description

@Bond-H

MoE模型使用DeepSpeed训练特别慢,GPU利用率不到30%

Megatron提供的示例应该是全参数微调的,需要的资源太大了,能否提供LoRA训练示例,这边改了下2张H20都跑不起来报OOM

Metadata

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

    Issue actions