[GH-ISSUE #267] How is the text encoder initialized? #233

Closed
opened 2026-10-31 21:12:46 +00:00 by bsakaguchi · 1 comment
Owner

Originally created by @sarahESL on GitHub (Jul 17, 2022).
Original GitHub issue: https://github.com/openai/CLIP/issues/267

The paper mentions that the text encoder is a Transformer with the architecture modifications from GPT-2.
My question is: is the text encoder trained from scratch or is it initialized by the learned parameters from GPT-2?

Thanks in advance for your answer.

Originally created by @sarahESL on GitHub (Jul 17, 2022). Original GitHub issue: https://github.com/openai/CLIP/issues/267 The paper mentions that the text encoder is a Transformer with the architecture modifications from GPT-2. My question is: is the text encoder trained from scratch or is it initialized by the learned parameters from GPT-2? Thanks in advance for your answer.
Author
Owner

@jongwook commented on GitHub (Nov 17, 2022):

It's trained from scratch.

<!-- gh-comment-id:1319083209 --> @jongwook commented on GitHub (Nov 17, 2022): It's trained from scratch.
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#233
No description provided.