[GH-ISSUE #3] Will there be code release for fine tuning or training? #1

Closed
opened 2026-10-31 21:09:45 +00:00 by bsakaguchi · 3 comments
Owner

Originally created by @sanmeow on GitHub (Jan 6, 2021).
Original GitHub issue: https://github.com/openai/CLIP/issues/3

Hi, this is a super great work!
Are there going to be anyway to fine tune this model for personal datasets?

Originally created by @sanmeow on GitHub (Jan 6, 2021). Original GitHub issue: https://github.com/openai/CLIP/issues/3 Hi, this is a super great work! Are there going to be anyway to fine tune this model for personal datasets?
Author
Owner

@jongwook commented on GitHub (Jan 7, 2021):

Thanks!

Fine-tuning should be possible with the released model. If you do the forward pass without the torch.no_grad() block, the resulting logits will have the gradients required for fine-tuning. You'll want to use the softmax cross entropy with the correct pairing (i.e. the identity matrix) as the target.

<!-- gh-comment-id:755789351 --> @jongwook commented on GitHub (Jan 7, 2021): Thanks! Fine-tuning should be possible with the released model. If you do the forward pass without the `torch.no_grad()` block, the resulting logits will have the gradients required for fine-tuning. You'll want to use the softmax cross entropy with the correct pairing (i.e. the identity matrix) as the target.
Author
Owner

@sanmeow commented on GitHub (Jan 7, 2021):

Thank!I will try it out!

<!-- gh-comment-id:755848746 --> @sanmeow commented on GitHub (Jan 7, 2021): Thank!I will try it out!
Author
Owner

@launchauto commented on GitHub (Jan 28, 2021):

Thank!I will try it out!

Do you success?

<!-- gh-comment-id:768767762 --> @launchauto commented on GitHub (Jan 28, 2021): > Thank!I will try it out! Do you success?
Sign in to join this conversation.
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
openai/CLIP#1
No description provided.