[GH-ISSUE #411] How do I fine-tune/train clip on mnist? #350

Open
opened 2026-10-31 21:14:14 +00:00 by bsakaguchi · 1 comment
Owner

Originally created by @GitOutOfMyBed on GitHub (Dec 16, 2023).
Original GitHub issue: https://github.com/openai/CLIP/issues/411

Does anyone know how to fine-tune clip on mnist? If I pass in a 32 batch of images and 10 unique labels, I don't know what the loss function would be.
Because in Clip's approach for each image, there's a unique text-image pairing. Now for my scenario, each image has a unique label but I do not have a unique image for each label.

Originally created by @GitOutOfMyBed on GitHub (Dec 16, 2023). Original GitHub issue: https://github.com/openai/CLIP/issues/411 Does anyone know how to fine-tune clip on mnist? If I pass in a 32 batch of images and 10 unique labels, I don't know what the loss function would be. Because in Clip's approach for each image, there's a unique text-image pairing. Now for my scenario, each image has a unique label but I do not have a unique image for each label.
Author
Owner

@tengshaofeng commented on GitHub (Feb 18, 2025):

https://github.com/tengshaofeng/finetune-jina-clip-v2

<!-- gh-comment-id:2665375768 --> @tengshaofeng commented on GitHub (Feb 18, 2025): https://github.com/tengshaofeng/finetune-jina-clip-v2
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#350
No description provided.