Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Full-stack PaliGemma #356

Closed
wants to merge 0 commits into from
Closed

Full-stack PaliGemma #356

wants to merge 0 commits into from

Conversation

copybara-service[bot]
Copy link

@copybara-service copybara-service bot commented Nov 13, 2024

Full-stack PaliGemma

  • Decoder should take embeddings combined with text and image instead of token ids.
  • If only text token ids are passed, call forward() as before.
  • If image embeddings are merged, position id is 1-based, not 0-based. RoPE calculation should take it into account.
  • If image embeddings are merged, diagonal causal mask doesn't work. Pass a mask just to remove unused KV entries.
  • Updated verifier API to get pixel_features
  • verifier.ReauthoredModelWrapper.generate() passes only new token in next round instead of all tokens of input + generated so far.
  • Conversion and unittests will be added in a following change.

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.

0 participants