Fix release full by ArthurZucker · Pull Request #45029 · huggingface/transformers

ArthurZucker · 2026-03-26T18:04:37Z

What does this PR do?

Release workflow is failing

HuggingFaceDocBuilderDev · 2026-03-26T18:20:41Z

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.

- `is_torchvision_available`, `is_timm_available`, `is_torchaudio_available`, `is_torchao_available`, `is_accelerate_available` now return False when torch is not installed, since all these packages require torch - Add `@requires(backends=("torch",))` to `PI0Processor` (was missing, causing the lazy module to crash on import without torch) - Fix wrong availability guards: `is_vision_available` → `is_torchvision_available` in pixtral processor, `is_torch_available` in smolvlm processor - Wrap bare `import torch` / torchvision imports in `processing_sam3_video.py` - Quote `torch.Tensor` in return type annotation of `tokenization_mistral_common.py` - Wrap 66 `image_processing_pil_*.py` imports from torch-dependent counterparts in try/except with ImagesKwargs fallbacks; quote `torch.Tensor` annotations - Restore explicit `from transformers import *` check in CircleCI `check_repository_consistency` job Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>

ArthurZucker · 2026-03-26T20:56:08Z

@vasqu could you review this? It needs to be fixed before we release — from transformers import * was broken without torch installed.

vasqu · 2026-03-26T22:24:15Z

src/transformers/utils/import_utils.py

+    if not is_torch_available():
+        return False


Is this an infinite loop? :D Edit: no sorry, its ao

@requires

The PIL image processor changes are too fragile (break on make fix-repo). Keep only the core fixes: - is_torchvision/timm/torchaudio/torchao/accelerate_available() check torch - CircleCI explicit import check - tokenization_mistral_common.py torch.Tensor annotation - processing_sam3_video.py conditional torch imports - processing_pixtral.py/processing_smolvlm.py availability guard fixes - PI0Processor @requires decorator Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>

github-actions · 2026-03-26T23:36:33Z

[For maintainers] Suggested jobs to run (before merge)

run-slow: aria, beit, bridgetower, conditional_detr, convnext, deepseek_vl, deepseek_vl_hybrid, deformable_detr, detr, donut, dpt, efficientloftr, efficientnet, eomt, ernie4_5_vl_moe, flava

ArthurZucker · 2026-03-26T22:44:01Z

src/transformers/utils/import_utils.py

-        "torch",
-        "torchvision",
-    ),
+    lambda name, content: name.startswith("image_processing_") and "pil" in name: ("vision",),


name.endswith("_fast"): now default don't

ArthurZucker · 2026-03-26T22:44:12Z

src/transformers/__init__.py

@@ -345,6 +345,7 @@
        name for name in dir(dummy_torchvision_objects) if not name.startswith("_")
    ]
 else:


torch vision depends on torch

second most important

ArthurZucker · 2026-03-26T23:46:36Z

src/transformers/utils/import_utils.py

+    lambda name, content: "tokenization_" in name and name.endswith("_fast"): ("tokenizers",),
+    lambda name, content: "image_processing_" in name and "TorchvisionBackend" in content: (
        "vision",
        "torch",
        "torchvision",
    ),
-    lambda name, content: name.startswith("image_processing_") and "TorchvisionBackend" in content: (
-        "vision",
-        "torch",
-        "torchvision",
-    ),
-    lambda name, content: name.startswith("image_processing_"): ("vision",),
-    lambda name, content: name.startswith("video_processing_"): ("vision", "torch", "torchvision"),
+    lambda name, content: "image_processing_" in name: ("vision",),
+    lambda name, content: "video_processing_" in name: ("vision", "torch", "torchvision"),


most important changes @LysandreJik this is why it was failing, name is full path

ArthurZucker · 2026-03-26T23:55:58Z

.circleci/config.yml

+            - run:
+                name: "Test import with all backends (torch + PIL + torchvision)"
+                command: python -c "from transformers import *" || (echo '🚨 import failed with all backends. Fix unprotected imports!! 🚨'; exit 1)
+            - run:
+                name: "Test import with torch only (no PIL, no torchvision)"
+                command: |
+                    uv pip uninstall Pillow torchvision -q
+                    python -c "from transformers import *" || (echo '🚨 import failed with torch only (no PIL). Fix unprotected imports!! 🚨'; exit 1)
+                    uv pip install -e ".[quality]" -q
+            - run:
+                name: "Test import with PIL only (no torch, no torchvision)"
+                command: |
+                    uv pip uninstall torch torchvision torchaudio -q
+                    python -c "from transformers import *" || (echo '🚨 import failed with PIL only (no torch). Fix unprotected imports!! 🚨'; exit 1)
+                    uv pip install -e ".[quality]" -q
+            - run:
+                name: "Test import with torch + PIL, no torchvision"
+                command: |
+                    uv pip uninstall torchvision -q
+                    python -c "from transformers import *" || (echo '🚨 import failed with torch+PIL but no torchvision. Fix unprotected imports!! 🚨'; exit 1)
+                    uv pip install -e ".[quality]" -q
            - run: make check-repository-consistency


cc @tarekziade and @ydshieh we are missing a bunch of these to enforce our import * is true WRT to our hards deps only

good catch, yeah the original check (moved to check-repository-consistency) only checked if transfomers is importable - we'll extend it

@requires

…age processors PR huggingface#45029 added @requires(backends=("vision", "torch", "torchvision")) to 67 PIL backend image_processing_pil_*.py files. This causes PIL backend classes to become dummy objects when torchvision is not installed, making AutoImageProcessor unable to find any working processor. Fix: set @requires to ("vision",) for files that only need PIL, and ("vision", "torch") for files that also use torch directly. Also fix 5 modular source files so make fix-repo preserves the correct backends. Fixes huggingface#45042 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

@requires

* first part of the fix * fix torch imports * revert * fix: make from transformers import * work without torch - `is_torchvision_available`, `is_timm_available`, `is_torchaudio_available`, `is_torchao_available`, `is_accelerate_available` now return False when torch is not installed, since all these packages require torch - Add `@requires(backends=("torch",))` to `PI0Processor` (was missing, causing the lazy module to crash on import without torch) - Fix wrong availability guards: `is_vision_available` → `is_torchvision_available` in pixtral processor, `is_torch_available` in smolvlm processor - Wrap bare `import torch` / torchvision imports in `processing_sam3_video.py` - Quote `torch.Tensor` in return type annotation of `tokenization_mistral_common.py` - Wrap 66 `image_processing_pil_*.py` imports from torch-dependent counterparts in try/except with ImagesKwargs fallbacks; quote `torch.Tensor` annotations - Restore explicit `from transformers import *` check in CircleCI `check_repository_consistency` job Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com> * up * style of this * revert: remove src/models changes, keep only core import fixes The PIL image processor changes are too fragile (break on make fix-repo). Keep only the core fixes: - is_torchvision/timm/torchaudio/torchao/accelerate_available() check torch - CircleCI explicit import check - tokenization_mistral_common.py torch.Tensor annotation - processing_sam3_video.py conditional torch imports - processing_pixtral.py/processing_smolvlm.py availability guard fixes - PI0Processor @requires decorator Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com> * nit * the mega quidproquo * use rquires(backend * more pil fixes * fixes * temp update * up? * is this it? * style? * revert a bunch of ai shit * pi0 requires this * revert some stuffs * upd * the fix * yups * ah * up * up * fix * yes? * update * up * nits * up * up * order --------- Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>

@requires

…age processors (#45045) * [Bugfix] Remove incorrect torchvision requirement from PIL backend image processors PR #45029 added @requires(backends=("vision", "torch", "torchvision")) to 67 PIL backend image_processing_pil_*.py files. This causes PIL backend classes to become dummy objects when torchvision is not installed, making AutoImageProcessor unable to find any working processor. Fix: set @requires to ("vision",) for files that only need PIL, and ("vision", "torch") for files that also use torch directly. Also fix 5 modular source files so make fix-repo preserves the correct backends. Fixes #45042 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * [Bugfix] Remove redundant @requires(backends=("vision",)) from PIL backends Per reviewer feedback: the vision-only @requires decorator is redundant for PIL backend classes since PilBackend base class already handles this. - Remove @requires(backends=("vision",)) from 43 PIL backend files - Remove unused `requires` import from 38 files (Category A) - Keep @requires(backends=("vision", "torch")) on method-level decorators (Category B: 5 files) * update * remove torch when its not necessary * remove if typechecking * fix import shinanigans * marvellous that's how we protect torch :) * beit is torchvisionbackend * more import cleanup * fiixup * fix-repo * update * style * fixes * up * more * fix repo * up * update * fix imports * style * fix check copies * arf * converter up * fix? * fix copies * fix for func * style * ignore * type --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Arthur <arthur.zucker@gmail.com>

@requires

* first part of the fix * fix torch imports * revert * fix: make from transformers import * work without torch - `is_torchvision_available`, `is_timm_available`, `is_torchaudio_available`, `is_torchao_available`, `is_accelerate_available` now return False when torch is not installed, since all these packages require torch - Add `@requires(backends=("torch",))` to `PI0Processor` (was missing, causing the lazy module to crash on import without torch) - Fix wrong availability guards: `is_vision_available` → `is_torchvision_available` in pixtral processor, `is_torch_available` in smolvlm processor - Wrap bare `import torch` / torchvision imports in `processing_sam3_video.py` - Quote `torch.Tensor` in return type annotation of `tokenization_mistral_common.py` - Wrap 66 `image_processing_pil_*.py` imports from torch-dependent counterparts in try/except with ImagesKwargs fallbacks; quote `torch.Tensor` annotations - Restore explicit `from transformers import *` check in CircleCI `check_repository_consistency` job Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com> * up * style of this * revert: remove src/models changes, keep only core import fixes The PIL image processor changes are too fragile (break on make fix-repo). Keep only the core fixes: - is_torchvision/timm/torchaudio/torchao/accelerate_available() check torch - CircleCI explicit import check - tokenization_mistral_common.py torch.Tensor annotation - processing_sam3_video.py conditional torch imports - processing_pixtral.py/processing_smolvlm.py availability guard fixes - PI0Processor @requires decorator Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com> * nit * the mega quidproquo * use rquires(backend * more pil fixes * fixes * temp update * up? * is this it? * style? * revert a bunch of ai shit * pi0 requires this * revert some stuffs * upd * the fix * yups * ah * up * up * fix * yes? * update * up * nits * up * up * order --------- Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>

@requires

…age processors (huggingface#45045) * [Bugfix] Remove incorrect torchvision requirement from PIL backend image processors PR huggingface#45029 added @requires(backends=("vision", "torch", "torchvision")) to 67 PIL backend image_processing_pil_*.py files. This causes PIL backend classes to become dummy objects when torchvision is not installed, making AutoImageProcessor unable to find any working processor. Fix: set @requires to ("vision",) for files that only need PIL, and ("vision", "torch") for files that also use torch directly. Also fix 5 modular source files so make fix-repo preserves the correct backends. Fixes huggingface#45042 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * [Bugfix] Remove redundant @requires(backends=("vision",)) from PIL backends Per reviewer feedback: the vision-only @requires decorator is redundant for PIL backend classes since PilBackend base class already handles this. - Remove @requires(backends=("vision",)) from 43 PIL backend files - Remove unused `requires` import from 38 files (Category A) - Keep @requires(backends=("vision", "torch")) on method-level decorators (Category B: 5 files) * update * remove torch when its not necessary * remove if typechecking * fix import shinanigans * marvellous that's how we protect torch :) * beit is torchvisionbackend * more import cleanup * fiixup * fix-repo * update * style * fixes * up * more * fix repo * up * update * fix imports * style * fix check copies * arf * converter up * fix? * fix copies * fix for func * style * ignore * type --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Arthur <arthur.zucker@gmail.com>

@requires

…age processors (huggingface#45045) * [Bugfix] Remove incorrect torchvision requirement from PIL backend image processors PR huggingface#45029 added @requires(backends=("vision", "torch", "torchvision")) to 67 PIL backend image_processing_pil_*.py files. This causes PIL backend classes to become dummy objects when torchvision is not installed, making AutoImageProcessor unable to find any working processor. Fix: set @requires to ("vision",) for files that only need PIL, and ("vision", "torch") for files that also use torch directly. Also fix 5 modular source files so make fix-repo preserves the correct backends. Fixes huggingface#45042 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * [Bugfix] Remove redundant @requires(backends=("vision",)) from PIL backends Per reviewer feedback: the vision-only @requires decorator is redundant for PIL backend classes since PilBackend base class already handles this. - Remove @requires(backends=("vision",)) from 43 PIL backend files - Remove unused `requires` import from 38 files (Category A) - Keep @requires(backends=("vision", "torch")) on method-level decorators (Category B: 5 files) * update * remove torch when its not necessary * remove if typechecking * fix import shinanigans * marvellous that's how we protect torch :) * beit is torchvisionbackend * more import cleanup * fiixup * fix-repo * update * style * fixes * up * more * fix repo * up * update * fix imports * style * fix check copies * arf * converter up * fix? * fix copies * fix for func * style * ignore * type --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Arthur <arthur.zucker@gmail.com>

ArthurZucker added 3 commits March 26, 2026 18:56

first part of the fix

93abcf9

fix torch imports

5d0c0c2

revert

2950cd2

vasqu reviewed Mar 26, 2026

View reviewed changes

ArthurZucker and others added 21 commits March 26, 2026 23:38

up

4028c51

style of this

500d73c

nit

e3c50a9

the mega quidproquo

ade0cd1

use rquires(backend

d0cab8e

more pil fixes

c463b18

fixes

93969d2

temp update

9feaf6f

up?

8545852

is this it?

e463e97

style?

9de946a

revert a bunch of ai shit

015b6b8

pi0 requires this

5bf5056

revert some stuffs

01d4a97

upd

f8ab9e1

the fix

a940f1a

yups

440b9de

ah

1509c73

up

c11d84f

up

9814fc4

fix

910801c

ArthurZucker added 3 commits March 27, 2026 00:38

yes?

81e1305

update

4b43926

up

f63fa89

ArthurZucker commented Mar 26, 2026

View reviewed changes

ArthurZucker marked this pull request as ready for review March 26, 2026 23:47

ArthurZucker added 3 commits March 27, 2026 00:50

nits

007bbdb

up

ef26d04

up

8669a89

ArthurZucker commented Mar 26, 2026

View reviewed changes

order

21f38de

ArthurZucker merged commit 97b7727 into main Mar 27, 2026
28 of 30 checks passed

ArthurZucker deleted the fix-release-full branch March 27, 2026 00:11

hysts mentioned this pull request Mar 27, 2026

PIL backend image processors incorrectly require torchvision in v5.4.0 #45042

Closed

4 tasks

Lidang-Jiang mentioned this pull request Mar 27, 2026

[Bugfix] Remove incorrect torchvision requirement from PIL backend image processors #45045

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix release full#45029

Fix release full#45029
ArthurZucker merged 33 commits intomainfrom
fix-release-full

ArthurZucker commented Mar 26, 2026

Uh oh!

HuggingFaceDocBuilderDev commented Mar 26, 2026

Uh oh!

ArthurZucker commented Mar 26, 2026

Uh oh!

vasqu Mar 26, 2026 •

edited

Loading

Uh oh!

github-actions bot commented Mar 26, 2026

Uh oh!

ArthurZucker Mar 26, 2026

Uh oh!

ArthurZucker Mar 26, 2026

Uh oh!

ArthurZucker Mar 26, 2026

Uh oh!

ArthurZucker Mar 26, 2026

Uh oh!

ArthurZucker Mar 26, 2026

Uh oh!

tarekziade Mar 27, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

ArthurZucker commented Mar 26, 2026

What does this PR do?

Uh oh!

HuggingFaceDocBuilderDev commented Mar 26, 2026

Uh oh!

ArthurZucker commented Mar 26, 2026

Uh oh!

vasqu Mar 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Mar 26, 2026

Uh oh!

ArthurZucker Mar 26, 2026

Choose a reason for hiding this comment

Uh oh!

ArthurZucker Mar 26, 2026

Choose a reason for hiding this comment

Uh oh!

ArthurZucker Mar 26, 2026

Choose a reason for hiding this comment

Uh oh!

ArthurZucker Mar 26, 2026

Choose a reason for hiding this comment

Uh oh!

ArthurZucker Mar 26, 2026

Choose a reason for hiding this comment

Uh oh!

tarekziade Mar 27, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

vasqu Mar 26, 2026 •

edited

Loading