Fix missing image processors backends#45165
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
yonigozlan
left a comment
yonigozlan
left a comment
There was a problem hiding this comment.
Hey @zucchini-nlp ! Sorry for the delay in reviewing. I'm actually not sure about this, as this is one case where we could have an issue with backward compatibility (only case where we removed an "image_processing_...py" without replacing it with a torchbackend, so if we try to import from transformers.models.emu3.image_processing_emu3 import Emu3ImageProcessor we will get an error).
I think we can leave it as it was until we add a torchvision backend for it. + in the current state, it's kind of a test to see if we support backward compatibility with old custom image processor (inheriting from BaseImageProcessor).
However the fix for blip is great thanks for catching this! Approving only for this part
|
Got it, thanks! I will add a small comment in Emu3 with reasons on why it is not updated then :) |
|
[For maintainers] Suggested jobs to run (before merge) run-slow: blip, emu3 |
* missing * revert emu3 * nope fix final test
* missing * revert emu3 * nope
* missing * revert emu3 * nope
3 participants
What does this PR do?
Emu3 was not updated in recent refactor and blip files were swapped. This PR fixes it
Do we need anything to support BC importing from old files, or does it happen in
LazyImports@yonigozlan ?