Conversation
|
The documentation is not available anymore as the PR was closed or merged. |
ArthurZucker
changed the title
dependency] update pillow pins
amyeroberts
left a comment
amyeroberts
left a comment
There was a problem hiding this comment.
Change LGTM. Could you run slow tests for some of the vision models before merging? I'd suggest: CLIP, ViT, DETR, SAM
setup.py
Outdated
| # 2. once modified, run: `make deps_table_update` to update src/transformers/dependency_versions_table.py | ||
| _deps = [ | ||
| "Pillow<10.0.0", | ||
| "Pillow>=10.0.1,<=11.0", |
There was a problem hiding this comment.
I'll pin <=15.0, it's just to not allow whatever!
Thanks for the tips will run the slow model tests 😉
There was a problem hiding this comment.
- Vit
- Clip
- Detr (tests/models/detr/test_modeling_detr.py::DetrModelIntegrationTestsTimmBackbone::test_inference_panoptic_segmentation_head fails:
E - {'id': 1, 'label_id': 17, 'score': 0.994097, 'was_fused': False}
E ? ^
E
E + {'id': 1, 'label_id': 17, 'score': 0.994096, 'was_fused': False}Might be my arch rather than anything
- SAM
Which models would you think rely a lot on pillow version that I should test as well?
There was a problem hiding this comment.
I think that difference is definitely in the realm of acceptable!
All vision models rely on Pillow for their processing. I picked these models to cover different default resampling algos + popular models on different tasks. AFAICT - there's just bicubic, bilinear and nearest used for resizing settings and they've all been tested :)
3 participants
What does this PR do?
Following, High Severity CVE, trying to update the pinned versions