Skip to content

Conversation

@bghira
Copy link
Owner

@bghira bghira commented Oct 26, 2025

we have to pass the tokenisers' original mask to T5 and only after, we build the Chroma pad mask. this mirrors upstream behaviour and keeps padding tokens from polluting the text embeds.

we'll append image-token masks using the existing mask dtype so that all types work without mismatch.

@bghira bghira merged commit 2dd050e into main Oct 26, 2025
1 check passed
@bghira bghira deleted the bugfix/chroma-masking-update branch October 26, 2025 17:17
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.

2 participants