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.

Sign up for GitHub

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

Jump to bottom

Improve SEG #1

Open
Abhinay1997 opened this issue Oct 8, 2024 · 1 comment
Open

Improve SEG #1

Abhinay1997 opened this issue Oct 8, 2024 · 1 comment

Comments

@Abhinay1997
Copy link
Owner

Abhinay1997 commented Oct 8, 2024 •
edited
Loading

  • Implementing SEG for both FluxTransformerBlock and FluxSingleTransformerBlock, only blurring the query of image tokens, not the text tokens.
  • Saving H and W of an image and using the aspect ratio instead of math.isqrt.
  • Applying SEG to a couple of blocks at around the 2/3 point of the array of transformer blocks.

Feedback from Susung Hong. Plan to implement it today.
@Abhinay1997
Copy link
Owner Author

Current implementaiton blurs all the tokens not just image ones and applies it to all the layers while ignoring the H & W and using math isqrt instead.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@Abhinay1997