Skip to content

Conversation

@65a
Copy link
Contributor

@65a 65a commented Nov 1, 2025

Not fully tested yet for vision (and maybe typos, so WIP)

Fixes #16908

@CISC
Copy link
Collaborator

CISC commented Nov 1, 2025

You seem to have inverted the embd size, anyway, mtmd will fail even if you correct this.

Try #16928

@65a
Copy link
Contributor Author

65a commented Nov 1, 2025

Dropping this one

@65a 65a closed this Nov 1, 2025
@65a
Copy link
Contributor Author

65a commented Nov 1, 2025

You seem to have inverted the embd size

I do want to understand this more: should it not equal the hidden state size of the text model?

@CISC
Copy link
Collaborator

CISC commented Nov 2, 2025

You seem to have inverted the embd size

I do want to understand this more: should it not equal the hidden state size of the text model?

Yes, but then you are using n_embd_full in the ubatch block.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Eval bug: Control vectors for Qwen3VL fail due to n_embd size

2 participants