multi-image training in GRPO #354

Open

opened

on Nov 25, 2025

During multi-image training in GRPO, I adjusted the model in the LORA section. When using both accuracy and format as rewards, the model failed to learn any content—it only learned the output format. Is there a solution to this problem?

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