Cosine-Similarity Loss is missing in the current code #3

Open

opened

on May 26, 2026

In the trainer.py which I assume is used for decoder + [cls] token training, only the flow loss is minimised and the cosine-similarity loss mentioned in the paper is missing. Why?

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