Fix: Add version check for timm to support mobilenetv5 models (fixes #39208) by VIGNESH15103 · Pull Request #39264 · huggingface/transformers

VIGNESH15103 · 2025-07-07T23:11:02Z

What does this PR do?

Fixes issue #39208: Adds a version check for timm to ensure support for models like mobilenetv5_300m_enc, which require timm >= 0.9.10.

Motivation

The issue occurred because the mobilenetv5_300m_enc model name was not recognized in older versions of timm. This PR adds a conditional check to raise an informative ImportError if an unsupported version is detected.

Changes

Added a version check using packaging.version inside TimmWrapperImageProcessor.__init__.
Raises an error if a mobilenetv5 model is used with timm < 0.9.10.

Issue Link

Closes #39208

Reviewer Suggestion

@amyeroberts @qubvel (vision models)

…uggingface#39208)

qubvel

Hey @VIGNESH15103, thanks a lot for the fix, I would suggest we add a check for config loading, not for the image processor

qubvel · 2025-07-08T09:39:44Z

+        # mobilenetv5 models require timm >= 0.9.10
+        if architecture is not None and architecture.startswith("mobilenetv5"):
+            if version.parse(timm.__version__) < version.parse("0.9.10"):
+                raise ImportError(
+                    f"`mobilenetv5` models require `timm >= 0.9.10`, but found {timm.__version__}. "
+                    "Please upgrade `timm` to the latest version."
+                )
+


The version for mibilenetv5 should be timm >= 1.0.16

Thanks @qubvel! Just to confirm — would it be appropriate to add the version check inside TimmWrapperModel.init and TimmWrapperForImageClassification.init right before timm.create_model(...) based on config.architecture? I’ll update accordingly.

Sound's good, but let's make it a bit more general, let's wrap timm.create_model to catch the RuntimeError and suggest to update timm with clear message

smt like

You are trying to instantiate `{architecture}`, but it's not supported in timm={timm.__version__}. Please, try updating to the latest timm version with `pip install -U timm`.

VIGNESH15103 · 2025-07-08T21:14:20Z

Following the feedback:

I removed the version check from image_processing_timm_wrapper.py.
Instead, I now catch RuntimeError during timm.create_model(...) inside both TimmWrapperModel and TimmWrapperForImageClassification, as suggested.
Clear ImportError is raised when the architecture is unsupported, prompting users to upgrade timm.
Applied ruff formatting to resolve CI failures.

Kindly requesting a review and workflow approval when convenient. Thanks a lot!

qubvel

Thanks for iterating, a few more comments 🤗

qubvel · 2025-07-10T08:47:47Z

+        except RuntimeError as e:
+            raise ImportError(
+                f"You are trying to instantiate `{config.architecture}`, but it's not supported in timm={timm.__version__}. "
+                "Please try updating to the latest timm version with `pip install -U timm`."
+            ) from e


We should catch an exact exception that the model is unsupported, not a general one

qubvel · 2025-07-10T08:48:04Z

+        # mobilenetv5 models require timm >= 0.9.10
+        if architecture is not None and architecture.startswith("mobilenetv5"):
+            if version.parse(timm.__version__) < version.parse("0.9.10"):
+                raise ImportError(
+                    f"`mobilenetv5` models require `timm >= 0.9.10`, but found {timm.__version__}. "
+                    "Please upgrade `timm` to the latest version."


do we still need this in image processor?

qubvel · 2025-07-10T08:48:44Z

+        try:
+            self.timm_model = timm.create_model(
+                config.architecture, pretrained=False, num_classes=config.num_labels, **extra_init_kwargs
+            )
+        except RuntimeError as e:
+            raise ImportError(
+                f"You are trying to instantiate `{config.architecture}`, but it's not supported in timm=={timm.__version__}. "
+                "Please try updating to the latest timm version with `pip install -U timm`."
+            ) from e
+


maybe wrap this in a function

github-actions · 2025-07-13T01:09:28Z

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

run-slow: timm_wrapper

…remove redundant checks

VIGNESH15103 · 2025-07-13T01:30:33Z

Hi! I've addressed all your feedback:

Wrapped model creation in a helper method as suggested.
Added specific exception handling.
Removed the redundant version check in the image processor.
Ran ruff formatting and all CI checks pass now.

Could you please take another look and let me know if there's anything else to improve? If everything looks good, it would be awesome if you could approve and merge it 😊

qubvel

Hey! I guess you are using some AI model to create PR, please review changes before submitting them 🤗

qubvel · 2025-07-14T11:13:04Z

+        # mobilenetv5 models require timm >= 0.9.10
+


qubvel · 2025-07-14T11:13:55Z

+        try:
+            self.timm_model = safe_create_model(config.architecture, 0, **extra_init_kwargs)
+
+        except RuntimeError as e:
+            raise ImportError(
+                f"You are trying to instantiate `{config.architecture}`, but it's not supported in timm={timm.__version__}. "
+                "Please try updating to the latest timm version with `pip install -U timm`."
+            ) from e


we should just call for safe_create_model with named args, no need for try catch anymore

qubvel · 2025-07-14T11:14:28Z

+def safe_create_model(architecture, num_classes, **kwargs):
+    try:
+        return timm.create_model(
+            architecture,
+            pretrained=False,
+            num_classes=num_classes,
+            **kwargs,
+        )
+    except RuntimeError as e:
+        raise ImportError(
+            f"You are trying to instantiate `{architecture}`, but it's not supported in timm=={timm.__version__}. "
+            "Please try updating to the latest timm version with `pip install -U timm`."
+        ) from e


we should check for a specific message in the exception, not just catch RuntimeError

VIGNESH15103 and others added 3 commits July 7, 2025 15:57

Fix: Add version check for timm to support mobilenetv5 models (fixes h…

517a479

…uggingface#39208)

style: fix ruff import order in image_processing_timm_wrapper.py

a414398

Merge branch 'main' into fix-unknown-mobilenetv5

14b51e0

qubvel reviewed Jul 8, 2025

View reviewed changes

VIGNESH15103 added 2 commits July 8, 2025 13:53

Raise ImportError when unsupported timm model is used

194746b

Apply ruff formatting to modeling_timm_wrapper.py

22a2cee

VIGNESH15103 force-pushed the fix-unknown-mobilenetv5 branch from 50ae8c5 to 22a2cee Compare July 8, 2025 21:02

Merge branch 'main' into fix-unknown-mobilenetv5

1c18f0c

Merge branch 'main' into fix-unknown-mobilenetv5

88366de

qubvel reviewed Jul 10, 2025

View reviewed changes

Refactor: Wrap model creation in helper, improve exception handling, …

5bfcdab

…remove redundant checks

VIGNESH15103 force-pushed the fix-unknown-mobilenetv5 branch from 6afd22c to 5bfcdab Compare July 13, 2025 01:17

Merge branch 'main' into fix-unknown-mobilenetv5

ceb9bf3

qubvel reviewed Jul 14, 2025

View reviewed changes

qubvel mentioned this pull request Jul 21, 2025

Unknown Model (mobilenetv5_300m_enc) when loading Gemma 3n #39208

Closed

4 tasks

This was referenced Apr 29, 2026

Cumulative feature and defect updates from recent Transformers PRs evalstate/transformers#42

Open

Cumulative defect fixes from recent Transformers PRs evalstate/transformers#43

Open

Conversation

VIGNESH15103 commented Jul 7, 2025

What does this PR do?

Motivation

Changes

Issue Link

Reviewer Suggestion

Uh oh!

qubvel left a comment • 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

VIGNESH15103 commented Jul 8, 2025

Uh oh!

qubvel left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Jul 13, 2025

Uh oh!

VIGNESH15103 commented Jul 13, 2025

Uh oh!

qubvel left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

qubvel Jul 14, 2025 • 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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

qubvel left a comment •

edited

Loading

qubvel Jul 14, 2025 •

edited

Loading