The size of tensor a (8) must match the size of tensor b (64) at non-singleton dimension

[pharmapsychotic]

The size of tensor a (8) must match the size of tensor b (64) at non-singleton dimension

[Dahvikiin]

Sorry i didn't see your message! currently i use transformers==4.29.2, but you can use previous versions, always higher than =4.25.1, and yes that commit (i made a fork for myself, i didn't want to unintentionally update the extension and have it break again). i had no problem. if you want to update transformers edit requirements_versions.txt (don't forget to restore it before updating AUTO1111 or whatever you are using).
you may also need to delete the folders "clip_interrogator" and "clip_interrogator-0.6.0.dist-info" in \venv\Lib\site-packages.

[Photodeus]

Likely a duplicate to #33

[Photodeus]

Also this, a closed issue
#21

[Dahvikiin]

use the previous commit 1ecedb5 which uses clip-interrogator 0.6.0.

[jcrsantiago]

I can't get it to work either. I have tried both the lastest version and 1ecedb5

[Dahvikiin]

interesting, in "venv\Libsite-packages" delete the folders (should be 2) called "clip_interrogator" and "clip_interrogator-0.6.0" (if clip_interrogator-0.5.4 exists, the same). Although it is also true that I use a more recent version of transformers.

[jcrsantiago]

Which version of transfers do you use? Are you using it with 1ecedb5

[ronbere]

Sorry i didn't see your message! currently i use transformers==4.29.2, but you can use previous versions, always higher than =4.25.1, and yes that commit (i made a fork for myself, i didn't want to unintentionally update the extension and have it break again). i had no problem. if you want to update transformers edit requirements_versions.txt (don't forget to restore it before updating AUTO1111 or whatever you are using). you may also need to delete the folders "clip_interrogator" and "clip_interrogator-0.6.0.dist-info" in \venv\Lib\site-packages.

same issue with your fork...Loading CLIP model...
Loaded CLIP model and data in 3.89 seconds.
The size of tensor a (8) must match the size of tensor b (64) at non-singleton dimension 0

[Grocel]

Error happens again as reported in issue #64.