chore(linter): fixes for rule 16 by tarekziade · Pull Request #46023 · huggingface/transformers

tarekziade · 2026-05-18T05:16:10Z

What does this PR do?

Adds support for rules 16

tarekziade · 2026-05-18T05:22:49Z

run-slow: albert, align, altclip, autoformer, bit, blt, bridgetower, chinese_clip, clap, clip, clipseg, clvp, colmodernvbert, colpali, colqwen2, conditional_detr

github-actions · 2026-05-18T05:24:07Z

Workflow Run ⚙️

This comment contains run-slow, running the specified jobs:

models: ["models/albert", "models/align", "models/altclip", "models/autoformer", "models/bit", "models/blt", "models/bridgetower", "models/chinese_clip", "models/clap", "models/clip", "models/clipseg", "models/clvp", "models/colmodernvbert", "models/colpali", "models/colqwen2", "models/conditional_detr"]
quantizations: []

HuggingFaceDocBuilderDev · 2026-05-18T05:27:39Z

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.

github-actions · 2026-05-18T05:43:07Z

CI Results

Workflow Run ⚙️

Commit Info

Context	Commit	Description
RUN	64943dac	workflow commit (merge commit)
PR	15c0ac4f	branch commit (from PR)
main	ad327c9b	base commit (on `main`)

✅ No failing test specific to this PR 🎉 👏 !

vasqu

I'm good with the current changes.

It kind of remains how we want to handle TRF 018: Imo, it would make sense to call this in any case - the performance loss is negligible for much cleaner modular paths in the future; any init already covered in the super can be overwritten anyway. But I'm not sure what @zucchini-nlp meant with the weird init patterns because they probably have been removed (by limiting the model scope) --> is there anything that would interfere? Maybe we could even remove some small overrides if they already do what the super call does

zucchini-nlp · 2026-05-27T08:34:29Z

    image_std = [1.0, 1.0, 1.0]
    size = None
    default_to_square = True
-    do_convert_rgb = True


why is it being removed btw? i see a few more deletes below for do pad, resize

I think that's TRF016 kicking in here, it's not referenced. Maybe we need more edge cases?

hmm indeed, prob they assumed it happens automatically outside prerpcessor, because that is how image processing works. I will pass by processing changes one-by-one to verify it can actually be deleted then

zucchini-nlp

Re init weights thing, i followed the slack thread with Cyril and prob misunderstood that init weights once via super will set the flag on the weight already. Since it can be init correctly even after calling super, I am fine with it

My only q is about deleted do_xxx in some processors. Maybe we can split into 3 PRs for each rule, imo we don't need to PR small subsets of models if we are happy with calling super().init_weights always

github-actions · 2026-05-28T12:07:31Z

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

run-slow: gemma4, glm46v, glm4v, glmga, kosmos2_5, llama4, mask2former, maskformer, minicpmv4_6, mllama, oneformer, perception_lm, phi4_multimodal, qwen2_vl, qwen3_vl, video_llama_3

tarekziade · 2026-05-28T12:12:10Z

splited out rules 17 and 18 on their own PR

#46260
#46259

tarekziade · 2026-05-28T12:15:27Z

run-slow: gemma4, glm46v, glm4v, glmga, kosmos2_5, llama4, mask2former, maskformer, minicpmv4_6, mllama, oneformer, perception_lm, phi4_multimodal, qwen2_vl, qwen3_vl, video_llama_3

github-actions · 2026-05-28T12:16:48Z

Workflow Run ⚙️

This comment contains run-slow, running the specified jobs:

models: ["models/gemma4", "models/glm46v", "models/glm4v", "models/glmga", "models/kosmos2_5", "models/llama4", "models/mask2former", "models/maskformer", "models/minicpmv4_6", "models/mllama", "models/oneformer", "models/perception_lm", "models/phi4_multimodal", "models/qwen2_vl", "models/qwen3_vl", "models/video_llama_3"]
quantizations: []

github-actions · 2026-05-28T12:50:50Z

CI Results

Workflow Run ⚙️

Commit Info

Context	Commit	Description
RUN	e615adb2	workflow commit (merge commit)
PR	6ed253df	branch commit (from PR)
main	bc8f70a9	base commit (on `main`)

⚠️ Model CI failed to report results

The test failure analysis could not be completed. Please check the workflow run for details.

zucchini-nlp

A few comments and lets' revert unrelated commits

zucchini-nlp · 2026-05-29T07:16:46Z

 +    tie_word_embeddings: bool = True
 ```

+### TRF016


i suppose the docs contains unrelated changes

changes in this file and rules.toml will be removed from the patch. we just want to make the code base rule-16-compatible so we can update mlinter to the next release in another patch

zucchini-nlp · 2026-05-29T07:22:52Z

zucchini-nlp · 2026-05-29T07:23:20Z

also not related

zucchini-nlp · 2026-05-29T07:26:25Z

    do_resize = True
    do_rescale = True
    do_normalize = True
-    do_convert_rgb = True


seeing all video processors don't use donvert_rgb. I suppose the contribs assumed it gets caught up internally

imo we have to keep the flag and make sure it is actually used. We need to squeeze it in here:

if do_convert_rgb: video = self.convert_to_rgb(video)

transformers/src/transformers/video_processing_utils.py

Lines 321 to 324 in 4557092

if device is not None:

video = video.to(device)

processed_videos.append(video)

zucchini-nlp · 2026-05-29T07:27:28Z

unrelated as well

zucchini-nlp · 2026-05-29T07:30:28Z

    image_mean = [0.5, 0.5, 0.5]
    image_std = [0.5, 0.5, 0.5]
    size = {"height": 336, "width": 336}
-    do_resize = True


can we TRF ignore it? imo it is meant to always resize and we can actually check if do_resize=False and raise an error that it cannot be False, otherwise the output pixel values cannot be tensorized

zucchini-nlp · 2026-05-29T07:34:12Z

    do_rescale = True
    do_normalize = True
    default_to_square = False
-    do_center_crop = False


can also delete crop_size = None, same for PIL backend

tarekziade requested review from vasqu and zucchini-nlp May 18, 2026 05:16

tarekziade self-assigned this May 18, 2026

zucchini-nlp reviewed May 18, 2026

View reviewed changes

Comment thread utils/modular_model_converter.py Outdated

hmellor reviewed May 18, 2026

View reviewed changes

Comment thread src/transformers/models/gemma4/modeling_gemma4.py Outdated

vasqu approved these changes May 18, 2026

View reviewed changes

Comment thread utils/rules.toml Outdated

Comment thread utils/rules.toml

tarekziade force-pushed the tarek-new-rules branch from aba4f5d to e381a40 Compare May 19, 2026 07:54

zucchini-nlp reviewed May 27, 2026

View reviewed changes

tarekziade added 3 commits May 28, 2026 13:42

added the 3 new rules

fd4c603

only do 30 models

4d944a5

fix new model

2035a0f

tarekziade changed the title ~~chore(linter): fix for rules 16-17-18~~ chore(linter): fix for rule 16 May 28, 2026

Split TRF017 and TRF018 from TRF016 branch

6ed253d

tarekziade force-pushed the tarek-new-rules branch from cf5ea7b to 6ed253d Compare May 28, 2026 12:06

zucchini-nlp reviewed May 29, 2026

View reviewed changes

tarekziade changed the title ~~chore(linter): fix for rule 16~~ chore(linter): fixes for rule 16 May 29, 2026

	if device is not None:
	video = video.to(device)

	processed_videos.append(video)

Conversation

tarekziade commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Uh oh!

tarekziade commented May 18, 2026

Uh oh!

github-actions Bot commented May 18, 2026

Uh oh!

Uh oh!

HuggingFaceDocBuilderDev commented May 18, 2026

Uh oh!

github-actions Bot commented May 18, 2026

CI Results

Commit Info

Uh oh!

Uh oh!

vasqu left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

zucchini-nlp May 27, 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

zucchini-nlp left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented May 28, 2026

Uh oh!

tarekziade commented May 28, 2026

Uh oh!

tarekziade commented May 28, 2026

Uh oh!

github-actions Bot commented May 28, 2026

Uh oh!

github-actions Bot commented May 28, 2026

CI Results

Commit Info

Uh oh!

zucchini-nlp 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!

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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

tarekziade commented May 18, 2026 •

edited

Loading

zucchini-nlp May 27, 2026 •

edited

Loading