8000 sequence parallel optimization for latest transformer · Issue #312 · volcengine/verl · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content
sequence parallel optimization for latest transformer #312
Closed
@eric-haibin-lin

Description

@eric-haibin-lin

verl does not yet support sequence parlallelism with the latest version of transformer > 0.48. The implementation is at verl/models/transformers/monkey_patch.py

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions

      0