Skip to content
This repository was archived by the owner on Feb 7, 2025. It is now read-only.

Conversation

@Ashayp31
Copy link
Contributor

@Ashayp31 Ashayp31 commented Feb 9, 2023

No description provided.

@Ashayp31 Ashayp31 linked an issue Feb 9, 2023 that may be closed by this pull request
@Ashayp31 Ashayp31 marked this pull request as draft February 9, 2023 11:35
Signed-off-by: Walter Hugo Lopez Pinaya <ianonimato@hotmail.com>
@Ashayp31
Copy link
Contributor Author

@danieltudosiu the output from the transformer when sampling new images seems to always give the same token. Do you have any idea why this may be the training loss seems good so I'm not too sure why this may be happening.

@danieltudosiu
Copy link
Contributor

@Ashayp31 sorry I did not implement this, I think @Warvito implemented it. But if I am to guess I would say the actual token selection from the output distribution is top 1 instead of using the full distribution.

@Ashayp31
Copy link
Contributor Author

@Ashayp31 sorry I did not implement this, I think @Warvito implemented it. But if I am to guess I would say the actual token selection from the output distribution is top 1 instead of using the full distribution.

@Ashayp31 Ashayp31 closed this Feb 16, 2023
@Ashayp31 Ashayp31 reopened this Feb 16, 2023
@Warvito Warvito marked this pull request as ready for review February 17, 2023 16:44
@Warvito Warvito merged commit 8076e22 into main Feb 17, 2023
@Warvito Warvito deleted the 106-add-tutorial-training-vqvaevqgan-+-autoregressivetransformer-in-2d branch February 17, 2023 16:45
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Add tutorial training VQVAE/VQGAN + AutoregressiveTransformer in 2D

4 participants