Skip to content

如何开启 Sequence Parallel 训练 #408

@jimmysue

Description

@jimmysue

对于 14B 的模型如果分辨率较大,序列长度较长,单存 FSDP 训练还是会现存不够,怎么开启 Sequence Parallel ,或者其他节省现存的方式训练?

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