Skip to content

Bump transformers-mlinter to 0.1.5 and clear the new findings - #48259

Merged
tarekziade merged 13 commits into
mainfrom
mlinter-0.1.5
Aug 28, 2026
Merged

Bump transformers-mlinter to 0.1.5 and clear the new findings#48259
tarekziade merged 13 commits into
mainfrom
mlinter-0.1.5

Conversation

@tarekziade

@tarekziade tarekziade commented Aug 24, 2026 •
edited by github-actions Bot
Loading

Copy link
Copy Markdown
Collaborator

CPU CI GPU run-slow

What does this PR do?

setup.py points transformers-mlinter at main ahead of the 0.1.5 release. The new rules surface 30 findings; this branch takes that to 0. make typing and make check-repo pass.

⚠️ Before merge: pin setup.py back to ==0.1.5 once released (make fix-repo propagates to dependency_versions_table.py).

Findings come from three upstream changes:

TRF058 — 4 sites fixed

esmfold2: three init buffers → nn.Buffer. The boundaries torch.linspace is inlined into the assignment on purpose — the modular converter reorders locals after the self.* writes that use them, which would NameError in the generated file.

step3p7: not an init declaration but a register_buffer inside init_weights re-materializing a non-persistent buffer Siglip already declares. Converted to init.copy(...), matching Siglip's own _init_weights and the rotary buffers in that same method.

TRF038 — 6 tests added, 1 bug fixed, 3 allowlisted

Five new test files (451 lines) covering 6 findings: dpr (both encoders via the mixin, plus a standalone reader class since DPRReaderTokenizer replaces call with (questions, titles, texts); decode_best_spans restored from fd6bc38), bert_legacy (BertTokenizerLegacy and whitespace_tokenize — the parts prophetnet's tests don't reach), qwen3_5, fnet, convbert.

roberta was a bug, not a missing test. transformers.RobertaTokenizerFast tokenized character-by-character:

RobertaTokenizerFast -> ['T','h','i','s','i','s','a','t','e','s','t']
RobertaTokenizer -> ['This','Ġis','Ġa','Ġtest']

#40936 converted 13 models to XxxTokenizerFast = XxxTokenizer and missed roberta, leaving a hand-written class whose backend was never built (pre_tokenizer was None). Public and autodoc'd, but nothing in the repo imported it — hence unnoticed. AutoTokenizer never routes there, so the fix is contained: delete the module, add the alias. Docs unchanged, as with the other aliased models.

Allowlisted: bart, mobilebert, squeezebert — their tokenization_*.py define no classes at all (BartTokenizer = RobertaTokenizer), and #40936 deleted their tests deliberately. Marked in-file as a placeholder for an upstream rule change, not a permanent exemption.

TRF009 — 6 fixed, 10 allowlisted

Six configs now resolve sibling configs through CONFIG_MAPPING with AutoConfig as the sub_configs marker, as the rule prescribes and 99 other composite configs already do: cohere_asr, esmfold2, gemma4_assistant, gemma4_unified_assistant, ovis2, sam3. Round-trips resolve the same concrete classes; 552 tests pass across the six.

Ten allowlisted, reasoning recorded in rules.toml. Two need a reviewer decision:

  • generation_.py (nemotron3_5_asr, nemotron_asr_streaming) — the rule points at modular files, but the converter has no generation_.py support. The rule checks a file kind whose remedy doesn't exist yet.
  • sam3 — AutoModel.from_config would cover the forward pass, but convert_sam3_to_hf.py:69 maps a weight to text_encoder.text_projection.weight, a key only CLIPTextModelWithProjection owns.

The rest are back-compat tokenizer shims where the import is the module's purpose. shieldgemma2 is the one genuinely fixable case left — the converter does support processing files, so it could move to a modular. Left as follow-up.

🤖 mlinter review state

tarekziade
tarekziade commented Aug 24, 2026
View reviewed changes
Comment thread utils/rules.toml Resolved
tarekziade
tarekziade commented Aug 24, 2026
View reviewed changes
Comment thread utils/rules.toml Resolved
github-actions[bot]
github-actions Bot reviewed Aug 24, 2026
View reviewed changes

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Model linter — first pass

transformers-mlinter found 30 item(s) in the model files this PR touches. These are structural conventions a maintainer would otherwise flag by hand.

This is automated and advisory — it does not block merging.

rule count what it checks
TRF009 16 No file in a model directory may import implementation code from another model package.
TRF038 10 Every modeling-family source file needs a matching test file under tests/models/.
TRF058 4 Buffers must be declared as nn.Buffer attributes, not registered with register_buffer().
26 item(s) outside this diff
  • src/transformers/models/bart/tokenization_bart.py:1 TRF038 — TRF038: no test file found at tests/models/bart/test_tokenization_bart.py for src/transformers/models/bart/tokenization_bart.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/bart/tokenization_bart.py:16 TRF009 — TRF009: tokenization_bart.py imports implementation code from roberta. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/bert/tokenization_bert_legacy.py:1 TRF038 — TRF038: no test file found at tests/models/bert/test_tokenization_bert_legacy.py for src/transformers/models/bert/tokenization_bert_legacy.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/cohere_asr/configuration_cohere_asr.py:20 TRF009 — TRF009: configuration_cohere_asr.py imports implementation code from parakeet. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/convbert/tokenization_convbert.py:1 TRF038 — TRF038: no test file found at tests/models/convbert/test_tokenization_convbert.py for src/transformers/models/convbert/tokenization_convbert.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/convbert/tokenization_convbert.py:16 TRF009 — TRF009: tokenization_convbert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/distilbert/tokenization_distilbert.py:16 TRF009 — TRF009: tokenization_distilbert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/dpr/tokenization_dpr.py:1 TRF038 — TRF038: no test file found at tests/models/dpr/test_tokenization_dpr.py for src/transformers/models/dpr/tokenization_dpr.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/dpr/tokenization_dpr_fast.py:1 TRF038 — TRF038: no test file found at tests/models/dpr/test_tokenization_dpr.py for src/transformers/models/dpr/tokenization_dpr_fast.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/esmfold2/configuration_esmfold2.py:20 TRF009 — TRF009: configuration_esmfold2.py imports implementation code from esmc. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/fnet/tokenization_fnet.py:1 TRF038 — TRF038: no test file found at tests/models/fnet/test_tokenization_fnet.py for src/transformers/models/fnet/tokenization_fnet.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/fnet/tokenization_fnet.py:17 TRF009 — TRF009: tokenization_fnet.py imports implementation code from albert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/gemma4_assistant/configuration_gemma4_assistant.py:21 TRF009 — TRF009: configuration_gemma4_assistant.py imports implementation code from gemma4. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/gemma4_unified_assistant/configuration_gemma4_unified_assistant.py:20 TRF009 — TRF009: configuration_gemma4_unified_assistant.py imports implementation code from gemma4_unified. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/mobilebert/tokenization_mobilebert.py:1 TRF038 — TRF038: no test file found at tests/models/mobilebert/test_tokenization_mobilebert.py for src/transformers/models/mobilebert/tokenization_mobilebert.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/mobilebert/tokenization_mobilebert.py:16 TRF009 — TRF009: tokenization_mobilebert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/nemotron3_5_asr/generation_nemotron3_5_asr.py:15 TRF009 — TRF009: generation_nemotron3_5_asr.py imports implementation code from nemotron_asr_streaming. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/nemotron_asr_streaming/generation_nemotron_asr_streaming.py:20 TRF009 — TRF009: generation_nemotron_asr_streaming.py imports implementation code from parakeet. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/ovis2/configuration_ovis2.py:20 TRF009 — TRF009: configuration_ovis2.py imports implementation code from qwen2. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/qwen3_5/tokenization_qwen3_5.py:1 TRF038 — TRF038: no test file found at tests/models/qwen3_5/test_tokenization_qwen3_5.py for src/transformers/models/qwen3_5/tokenization_qwen3_5.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/roberta/tokenization_roberta_old.py:1 TRF038 — TRF038: no test file found at tests/models/roberta/test_tokenization_roberta_old.py for src/transformers/models/roberta/tokenization_roberta_old.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/sam3/configuration_sam3.py:18 TRF009 — TRF009: configuration_sam3.py imports implementation code from clip. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/sam3/modeling_sam3.py:31 TRF009 — TRF009: modeling_sam3.py imports implementation code from clip. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/shieldgemma2/processing_shieldgemma2.py:21 TRF009 — TRF009: processing_shieldgemma2.py imports implementation code from gemma3. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/squeezebert/tokenization_squeezebert.py:1 TRF038 — TRF038: no test file found at tests/models/squeezebert/test_tokenization_squeezebert.py for src/transformers/models/squeezebert/tokenization_squeezebert.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/squeezebert/tokenization_squeezebert.py:16 TRF009 — TRF009: tokenization_squeezebert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.

github-actions[bot]
github-actions Bot reviewed Aug 24, 2026
View reviewed changes

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Model linter — first pass

transformers-mlinter found 26 item(s) in the model files this PR touches. These are structural conventions a maintainer would otherwise flag by hand.

This is automated and advisory — it does not block merging.

rule count what it checks
TRF009 16 No file in a model directory may import implementation code from another model package.
TRF038 10 Every modeling-family source file needs a matching test file under tests/models/.
26 item(s) outside this diff
  • src/transformers/models/bart/tokenization_bart.py:1 TRF038 — TRF038: no test file found at tests/models/bart/test_tokenization_bart.py for src/transformers/models/bart/tokenization_bart.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/bart/tokenization_bart.py:16 TRF009 — TRF009: tokenization_bart.py imports implementation code from roberta. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/bert/tokenization_bert_legacy.py:1 TRF038 — TRF038: no test file found at tests/models/bert/test_tokenization_bert_legacy.py for src/transformers/models/bert/tokenization_bert_legacy.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/cohere_asr/configuration_cohere_asr.py:20 TRF009 — TRF009: configuration_cohere_asr.py imports implementation code from parakeet. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/convbert/tokenization_convbert.py:1 TRF038 — TRF038: no test file found at tests/models/convbert/test_tokenization_convbert.py for src/transformers/models/convbert/tokenization_convbert.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/convbert/tokenization_convbert.py:16 TRF009 — TRF009: tokenization_convbert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/distilbert/tokenization_distilbert.py:16 TRF009 — TRF009: tokenization_distilbert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/dpr/tokenization_dpr.py:1 TRF038 — TRF038: no test file found at tests/models/dpr/test_tokenization_dpr.py for src/transformers/models/dpr/tokenization_dpr.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/dpr/tokenization_dpr_fast.py:1 TRF038 — TRF038: no test file found at tests/models/dpr/test_tokenization_dpr.py for src/transformers/models/dpr/tokenization_dpr_fast.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/esmfold2/configuration_esmfold2.py:20 TRF009 — TRF009: configuration_esmfold2.py imports implementation code from esmc. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/fnet/tokenization_fnet.py:1 TRF038 — TRF038: no test file found at tests/models/fnet/test_tokenization_fnet.py for src/transformers/models/fnet/tokenization_fnet.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/fnet/tokenization_fnet.py:17 TRF009 — TRF009: tokenization_fnet.py imports implementation code from albert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/gemma4_assistant/configuration_gemma4_assistant.py:21 TRF009 — TRF009: configuration_gemma4_assistant.py imports implementation code from gemma4. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/gemma4_unified_assistant/configuration_gemma4_unified_assistant.py:20 TRF009 — TRF009: configuration_gemma4_unified_assistant.py imports implementation code from gemma4_unified. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/mobilebert/tokenization_mobilebert.py:1 TRF038 — TRF038: no test file found at tests/models/mobilebert/test_tokenization_mobilebert.py for src/transformers/models/mobilebert/tokenization_mobilebert.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/mobilebert/tokenization_mobilebert.py:16 TRF009 — TRF009: tokenization_mobilebert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/nemotron3_5_asr/generation_nemotron3_5_asr.py:15 TRF009 — TRF009: generation_nemotron3_5_asr.py imports implementation code from nemotron_asr_streaming. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/nemotron_asr_streaming/generation_nemotron_asr_streaming.py:20 TRF009 — TRF009: generation_nemotron_asr_streaming.py imports implementation code from parakeet. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/ovis2/configuration_ovis2.py:20 TRF009 — TRF009: configuration_ovis2.py imports implementation code from qwen2. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/qwen3_5/tokenization_qwen3_5.py:1 TRF038 — TRF038: no test file found at tests/models/qwen3_5/test_tokenization_qwen3_5.py for src/transformers/models/qwen3_5/tokenization_qwen3_5.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/roberta/tokenization_roberta_old.py:1 TRF038 — TRF038: no test file found at tests/models/roberta/test_tokenization_roberta_old.py for src/transformers/models/roberta/tokenization_roberta_old.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/sam3/configuration_sam3.py:18 TRF009 — TRF009: configuration_sam3.py imports implementation code from clip. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/sam3/modeling_sam3.py:31 TRF009 — TRF009: modeling_sam3.py imports implementation code from clip. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/shieldgemma2/processing_shieldgemma2.py:21 TRF009 — TRF009: processing_shieldgemma2.py imports implementation code from gemma3. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/squeezebert/tokenization_squeezebert.py:1 TRF038 — TRF038: no test file found at tests/models/squeezebert/test_tokenization_squeezebert.py for src/transformers/models/squeezebert/tokenization_squeezebert.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/squeezebert/tokenization_squeezebert.py:16 TRF009 — TRF009: tokenization_squeezebert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.

@HuggingFaceDocBuilderDev

HuggingFaceDocBuilderDev commented Aug 24, 2026

Copy link
Copy Markdown

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[bot]
github-actions Bot reviewed Aug 24, 2026
View reviewed changes

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Model linter — first pass

transformers-mlinter found 20 item(s) in the model files this PR touches. These are structural conventions a maintainer would otherwise flag by hand.

This is automated and advisory — it does not block merging.

rule count what it checks
TRF009 16 No file in a model directory may import implementation code from another model package.
TRF038 4 Every modeling-family source file needs a matching test file under tests/models/.
20 item(s) outside this diff
  • src/transformers/models/bart/tokenization_bart.py:1 TRF038 — TRF038: no test file found at tests/models/bart/test_tokenization_bart.py for src/transformers/models/bart/tokenization_bart.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/bart/tokenization_bart.py:16 TRF009 — TRF009: tokenization_bart.py imports implementation code from roberta. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/cohere_asr/configuration_cohere_asr.py:20 TRF009 — TRF009: configuration_cohere_asr.py imports implementation code from parakeet. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/convbert/tokenization_convbert.py:16 TRF009 — TRF009: tokenization_convbert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/distilbert/tokenization_distilbert.py:16 TRF009 — TRF009: tokenization_distilbert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/esmfold2/configuration_esmfold2.py:20 TRF009 — TRF009: configuration_esmfold2.py imports implementation code from esmc. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/fnet/tokenization_fnet.py:17 TRF009 — TRF009: tokenization_fnet.py imports implementation code from albert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/gemma4_assistant/configuration_gemma4_assistant.py:21 TRF009 — TRF009: configuration_gemma4_assistant.py imports implementation code from gemma4. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/gemma4_unified_assistant/configuration_gemma4_unified_assistant.py:20 TRF009 — TRF009: configuration_gemma4_unified_assistant.py imports implementation code from gemma4_unified. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/mobilebert/tokenization_mobilebert.py:1 TRF038 — TRF038: no test file found at tests/models/mobilebert/test_tokenization_mobilebert.py for src/transformers/models/mobilebert/tokenization_mobilebert.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/mobilebert/tokenization_mobilebert.py:16 TRF009 — TRF009: tokenization_mobilebert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/nemotron3_5_asr/generation_nemotron3_5_asr.py:15 TRF009 — TRF009: generation_nemotron3_5_asr.py imports implementation code from nemotron_asr_streaming. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/nemotron_asr_streaming/generation_nemotron_asr_streaming.py:20 TRF009 — TRF009: generation_nemotron_asr_streaming.py imports implementation code from parakeet. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/ovis2/configuration_ovis2.py:20 TRF009 — TRF009: configuration_ovis2.py imports implementation code from qwen2. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/roberta/tokenization_roberta_old.py:1 TRF038 — TRF038: no test file found at tests/models/roberta/test_tokenization_roberta_old.py for src/transformers/models/roberta/tokenization_roberta_old.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/sam3/configuration_sam3.py:18 TRF009 — TRF009: configuration_sam3.py imports implementation code from clip. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/sam3/modeling_sam3.py:31 TRF009 — TRF009: modeling_sam3.py imports implementation code from clip. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/shieldgemma2/processing_shieldgemma2.py:21 TRF009 — TRF009: processing_shieldgemma2.py imports implementation code from gemma3. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/squeezebert/tokenization_squeezebert.py:1 TRF038 — TRF038: no test file found at tests/models/squeezebert/test_tokenization_squeezebert.py for src/transformers/models/squeezebert/tokenization_squeezebert.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/squeezebert/tokenization_squeezebert.py:16 TRF009 — TRF009: tokenization_squeezebert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.

github-actions[bot]
github-actions Bot reviewed Aug 24, 2026
View reviewed changes

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Model linter — first pass

transformers-mlinter found 19 item(s) in the model files this PR touches. These are structural conventions a maintainer would otherwise flag by hand.

This is automated and advisory — it does not block merging.

rule count what it checks
TRF009 16 No file in a model directory may import implementation code from another model package.
TRF038 3 Every modeling-family source file needs a matching test file under tests/models/.
19 item(s) outside this diff
  • src/transformers/models/bart/tokenization_bart.py:1 TRF038 — TRF038: no test file found at tests/models/bart/test_tokenization_bart.py for src/transformers/models/bart/tokenization_bart.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/bart/tokenization_bart.py:16 TRF009 — TRF009: tokenization_bart.py imports implementation code from roberta. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/cohere_asr/configuration_cohere_asr.py:20 TRF009 — TRF009: configuration_cohere_asr.py imports implementation code from parakeet. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/convbert/tokenization_convbert.py:16 TRF009 — TRF009: tokenization_convbert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/distilbert/tokenization_distilbert.py:16 TRF009 — TRF009: tokenization_distilbert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/esmfold2/configuration_esmfold2.py:20 TRF009 — TRF009: configuration_esmfold2.py imports implementation code from esmc. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/fnet/tokenization_fnet.py:17 TRF009 — TRF009: tokenization_fnet.py imports implementation code from albert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/gemma4_assistant/configuration_gemma4_assistant.py:21 TRF009 — TRF009: configuration_gemma4_assistant.py imports implementation code from gemma4. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/gemma4_unified_assistant/configuration_gemma4_unified_assistant.py:20 TRF009 — TRF009: configuration_gemma4_unified_assistant.py imports implementation code from gemma4_unified. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/mobilebert/tokenization_mobilebert.py:1 TRF038 — TRF038: no test file found at tests/models/mobilebert/test_tokenization_mobilebert.py for src/transformers/models/mobilebert/tokenization_mobilebert.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/mobilebert/tokenization_mobilebert.py:16 TRF009 — TRF009: tokenization_mobilebert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/nemotron3_5_asr/generation_nemotron3_5_asr.py:15 TRF009 — TRF009: generation_nemotron3_5_asr.py imports implementation code from nemotron_asr_streaming. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/nemotron_asr_streaming/generation_nemotron_asr_streaming.py:20 TRF009 — TRF009: generation_nemotron_asr_streaming.py imports implementation code from parakeet. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/ovis2/configuration_ovis2.py:20 TRF009 — TRF009: configuration_ovis2.py imports implementation code from qwen2. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/sam3/configuration_sam3.py:18 TRF009 — TRF009: configuration_sam3.py imports implementation code from clip. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/sam3/modeling_sam3.py:31 TRF009 — TRF009: modeling_sam3.py imports implementation code from clip. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/shieldgemma2/processing_shieldgemma2.py:21 TRF009 — TRF009: processing_shieldgemma2.py imports implementation code from gemma3. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/squeezebert/tokenization_squeezebert.py:1 TRF038 — TRF038: no test file found at tests/models/squeezebert/test_tokenization_squeezebert.py for src/transformers/models/squeezebert/tokenization_squeezebert.py. Add one, even a minimal test built on a small hand-written vocabulary if no real checkpoint exists yet.
  • src/transformers/models/squeezebert/tokenization_squeezebert.py:16 TRF009 — TRF009: tokenization_squeezebert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.

github-actions[bot]
github-actions Bot reviewed Aug 24, 2026
View reviewed changes

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Model linter — first pass

transformers-mlinter found 16 item(s) in the model files this PR touches. These are structural conventions a maintainer would otherwise flag by hand.

This is automated and advisory — it does not block merging.

rule count what it checks
TRF009 16 No file in a model directory may import implementation code from another model package.
16 item(s) outside this diff
  • src/transformers/models/bart/tokenization_bart.py:16 TRF009 — TRF009: tokenization_bart.py imports implementation code from roberta. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/cohere_asr/configuration_cohere_asr.py:20 TRF009 — TRF009: configuration_cohere_asr.py imports implementation code from parakeet. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/convbert/tokenization_convbert.py:16 TRF009 — TRF009: tokenization_convbert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/distilbert/tokenization_distilbert.py:16 TRF009 — TRF009: tokenization_distilbert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/esmfold2/configuration_esmfold2.py:20 TRF009 — TRF009: configuration_esmfold2.py imports implementation code from esmc. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/fnet/tokenization_fnet.py:17 TRF009 — TRF009: tokenization_fnet.py imports implementation code from albert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/gemma4_assistant/configuration_gemma4_assistant.py:21 TRF009 — TRF009: configuration_gemma4_assistant.py imports implementation code from gemma4. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/gemma4_unified_assistant/configuration_gemma4_unified_assistant.py:20 TRF009 — TRF009: configuration_gemma4_unified_assistant.py imports implementation code from gemma4_unified. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/mobilebert/tokenization_mobilebert.py:16 TRF009 — TRF009: tokenization_mobilebert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/nemotron3_5_asr/generation_nemotron3_5_asr.py:15 TRF009 — TRF009: generation_nemotron3_5_asr.py imports implementation code from nemotron_asr_streaming. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/nemotron_asr_streaming/generation_nemotron_asr_streaming.py:20 TRF009 — TRF009: generation_nemotron_asr_streaming.py imports implementation code from parakeet. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/ovis2/configuration_ovis2.py:20 TRF009 — TRF009: configuration_ovis2.py imports implementation code from qwen2. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/sam3/configuration_sam3.py:18 TRF009 — TRF009: configuration_sam3.py imports implementation code from clip. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/sam3/modeling_sam3.py:31 TRF009 — TRF009: modeling_sam3.py imports implementation code from clip. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/shieldgemma2/processing_shieldgemma2.py:21 TRF009 — TRF009: processing_shieldgemma2.py imports implementation code from gemma3. Keep model code local to this model's own files; only modular_*.py may build on another model.
  • src/transformers/models/squeezebert/tokenization_squeezebert.py:16 TRF009 — TRF009: tokenization_squeezebert.py imports implementation code from bert. Keep model code local to this model's own files; only modular_*.py may build on another model.

@tarekziade

tarekziade commented Aug 24, 2026

Copy link
Copy Markdown
Collaborator Author

run-slow: bert, cohere_asr, convbert, dpr, esmfold2, fnet, gemma4_assistant, gemma4_unified_assistant, ovis2, qwen3_5, roberta, sam3, step3p7

@github-actions

github-actions Bot commented Aug 24, 2026

Copy link
Copy Markdown
Contributor

Workflow Run ⚙️

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

models: ["models/bert", "models/cohere_asr", "models/convbert", "models/dpr", "models/esmfold2", "models/fnet", "models/gemma4_assistant", "models/gemma4_unified_assistant", "models/ovis2", "models/qwen3_5", "models/roberta", "models/sam3", "models/step3p7"]
quantizations: []

@github-actions

github-actions Bot commented Aug 24, 2026

Copy link
Copy Markdown
Contributor

CI Results

Workflow Run ⚙️

Commit Info

Context Commit Description
RUN 8338db28 workflow commit (merge commit)
PR da722775 branch commit (from PR)
main a530903e base commit (on main)

✅ No failing test specific to this PR 🎉 👏 !

@tarekziade tarekziade changed the title future mlinter 0.1.5 Bump transformers-mlinter to 0.1.5 and clear the new findings Aug 24, 2026
@tarekziade tarekziade self-assigned this Aug 24, 2026
zucchini-nlp
zucchini-nlp approved these changes Aug 25, 2026
View reviewed changes

@zucchini-nlp zucchini-nlp left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yep, lgtm! Not sure if we're bumping it now since I saw an internal slack thread with bugfix requests

Approving in advance so it can be merged whenever we're ready

Comment thread src/transformers/models/cohere_asr/configuration_cohere_asr.py
from ...utils import auto_docstring
from ..auto import CONFIG_MAPPING
from ..parakeet.configuration_parakeet import ParakeetEncoderConfig
from ..auto import CONFIG_MAPPING, AutoConfig

@zucchini-nlp zucchini-nlp Aug 25, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

omg, didn't know we had this many of cross-imports 😢

Comment thread src/transformers/models/roberta/tokenization_roberta.py

__all__ = ["RobertaTokenizer"]
# RobertaTokenizerFast is an alias for RobertaTokenizer (since RobertaTokenizer is already a fast tokenizer)
RobertaTokenizerFast = RobertaTokenizer

@zucchini-nlp zucchini-nlp Aug 25, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Huh interesting, not sure what is the best way to keep a single version after Ita's tokenization backend refactor. Ig we can keep both for BC indeed

Comment thread tests/models/bert/test_tokenization_bert_legacy.py
Comment on lines +14 to +24
"""
Tests for `BertTokenizerLegacy`, the pure-Python WordPiece tokenizer.

It is not reachable through `AutoTokenizer` any more, but it is far from unused: `pipelines/token_classification.py`
imports `BasicTokenizer` from this module, `data/processors/squad.py` imports `whitespace_tokenize`, and the
tokenizers of tapas, roc_bert, prophetnet, bert_japanese and openai are all built on its helpers.

The `BasicTokenizer` / `WordpieceTokenizer` casing and accent matrix is exercised in
`tests/models/prophetnet/test_tokenization_prophetnet.py`, which imports those helpers from this module. What is
covered here is the part nothing else reaches: the `BertTokenizerLegacy` class itself, and `whitespace_tokenize`.
"""

@zucchini-nlp zucchini-nlp Aug 25, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nice!

Comment thread utils/rules.toml
Comment on lines +150 to +155
# Back-compat tokenizer shims: `bart`, `convbert`, `distilbert`, `fnet`, `mobilebert`, `squeezebert`. Each
# tokenization_*.py exists only to re-export or thinly subclass another model's tokenizer, so the import is
# the module's entire purpose -- `BartTokenizer = RobertaTokenizer` cannot be written without importing
# RobertaTokenizer. The alternatives are deleting a public name or copying a whole WordPiece/BPE
# implementation into six files. Same population as the TRF038 entries below.
#

@zucchini-nlp zucchini-nlp Aug 25, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ah, good to know! I suppose it is only for BC (when users import directly BartTokenizer, etc) since they can be auto-mapped to correctly load classes

Comment thread utils/rules.toml
# mixin and cache, which the rule says belongs in a modular file -- but the modular converter has no
# generation_*.py support (no generated one exists in the repo, and the converter never mentions the file
# kind), so the prescribed remedy is unavailable. Either the converter learns this file kind or the rule
# stops checking it; flagged upstream rather than worked around here.

@zucchini-nlp zucchini-nlp Aug 25, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

flagged upstream rather than worked around here - thanks, imo this is important to fix in modular conversion

Comment thread utils/rules.toml Outdated
what_it_does = "Checks modeling files for cross-model imports such as transformers.models.other_model.* or from ..other_model.* imports."
why_bad = "Cross-model implementation imports violate the single-file policy and make model behavior harder to inspect and maintain."
what_it_does = "Flags imports reaching into another model's package, in the files that make up a model's shipped implementation: modeling_*, configuration_*, processing_*, image_processing_*, video_processing_*, feature_extraction_*, tokenization_* and generation_*.py. Three forms are recognised: package path (`from transformers.models.other.modeling_other import X`), relative path (`from ..other.configuration_other import X`), and public API (`from transformers import OtherModel`) -- for the last, the owning directory is inferred from the class name and confirmed against the classes that directory really defines, so an unresolvable name is left alone rather than guessed at. Out of scope: `modular_*.py` (building on another model is its purpose, and the converter flattens those imports away), `convert_*.py`, `__init__.py` and files under `auto`. Allowed targets: the model's own directory, `auto` and `timm_wrapper`."
why_bad = "One model, one definition: all of Acme's behavior should live in Acme's own files, and a change to Llama must not silently change Acme. This holds for every file kind, not just the modeling one. Reach sub-configs through AutoConfig and CONFIG_MAPPING, and use a modular file to build on another model's code."

@zucchini-nlp zucchini-nlp Aug 25, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: why Acme, should we say "model"? 😆

@tarekziade tarekziade Aug 28, 2026

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

haha

ArthurZucker
ArthurZucker approved these changes Aug 25, 2026
View reviewed changes

@ArthurZucker ArthurZucker left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm

tarekziade and others added 11 commits August 28, 2026 06:21
@tarekziade
future mlinter 0.1.5
f64ff29
@tarekziade
fix rule 58
93f7089
@tarekziade
Add tokenizer tests for TRF038
e619388 
Adds the missing tokenizer test files that mlinter's TRF038 reports, now that
the rule covers tokenization_*.py files.
@tarekziade
fix roberta
25a5260
@tarekziade
Allowlist the three alias-only tokenizer modules for TRF038
cc7f1ae 
bart, mobilebert and squeezebert started firing when mlinter 0.1.5 extended
TRF038 to tokenizer files. Their tokenization_*.py define no classes at all,
each being a two-line back-compat alias left by "rm slow tokenizers" (#40936),
which deleted their test files on purpose. A test file here could only assert
alias identity, and the aliased tokenizers are already covered by
tests/models/roberta and tests/models/bert.

Allowlisted as a placeholder: the better fix is upstream, where TRF038 should
skip a source file that defines no classes.

TRF038 now reports no findings.
@tarekziade @claude
Route composite sub-configs through CONFIG_MAPPING for TRF009
447933e 
mlinter 0.1.5 extended TRF009 from modeling files to every model file, which
surfaced six configs importing a sibling model's config class directly. Each now
resolves it through CONFIG_MAPPING with AutoConfig as the sub_configs marker,
which is what the rule prescribes and what 99 other composite configs already
do:

  cohere_asr               ParakeetEncoderConfig    -> parakeet_encoder
  esmfold2                 EsmcConfig               -> esmc
  gemma4_assistant         Gemma4TextConfig         -> gemma4_text
  gemma4_unified_assistant Gemma4UnifiedTextConfig  -> gemma4_unified_text
  ovis2                    Qwen2Config              -> qwen2
  sam3                     CLIPTextConfig           -> clip_text_model

cohere_asr already used CONFIG_MAPPING for instantiation, so only its
sub_configs value and import changed. Where the imported class also appeared in
a type annotation it is widened to PreTrainedConfig, and esmfold2's _init_nested
helper takes the class from CONFIG_MAPPING since it calls cls() directly.

Verified: default construction and to_dict/from_dict round-trip resolve the same
concrete config class as before for all six, and tests/models for the six models
pass (552 passed, 551 skipped).

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@tarekziade
Backport mlinter rule-prose compression and allowlist the rest of TRF009
ffbdaeb 
Two changes to utils/rules.toml.

Backports transformers-mlinter d0ac218 ("compress the rule explanation prose",
before: upstream's file as the base, with this repo's licence header and its
allowlist_models entries and their justification comments re-applied on top, so
the only remaining difference from the installed mlinter/rules.toml is the header
and those allowlists. make fix-repo regenerated
docs/source/en/modeling_rules.md.

Allowlists the ten TRF009 findings left after the CONFIG_MAPPING fixes, in three
groups, each with its reasoning recorded in the file:

- Back-compat tokenizer shims (bart, convbert, distilbert, fnet, mobilebert,
  squeezebert): the cross-model import is the module's whole purpose.
- generation_*.py (nemotron3_5_asr, nemotron_asr_streaming): the rule points at
  modular files, but the converter has no generation_*.py support, so the
  prescribed remedy does not exist yet.
- sam3: CLIPTextModelWithProjection is needed for checkpoint key compatibility.
  shieldgemma2: fixable with a modular file, left as follow-up.

make typing and make check-repo now pass.
@tarekziade @claude
Fix vibevoice for mlinter 0.1.5 after rebase
3ad6246 
VibeVoice (#40546) landed on main after this branch forked, so it was never
checked against the 0.1.5 rules and brought in two findings.

TRF058: VibeVoiceDiffusionHeadSinusoidalEmbedding created `freq` via
`self.register_buffer(...)`, which cannot be inherited and tweaked from a
modular file. Assign `nn.Buffer(..., persistent=False)` instead, which also
makes the `freq: torch.Tensor` linting workaround unnecessary. Same
non-persistent buffer and same forward output.

TRF009: allowlist `vibevoice`. generation_vibevoice.py imports a runtime cache
class from its own sub-model package; the family already routes its sub-configs
through CONFIG_MAPPING, and a cache class cannot come from CONFIG_MAPPING. It
also sits in a generation_*.py file, which the modular converter does not
support, the same blocker as the nemotron entries.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@tarekziade @claude
Adopt transformers-mlinter main: sync rule prose, drop dead allowlist…
5a1b82c 
… entries

Re-ran the checker against transformers-mlinter git main (e6533da). Green, and
the bump makes a chunk of this branch's curation obsolete.

Synced the rule prose the repo copy mirrors, which had gone stale because the
described behavior changed upstream:
  TRF029 -- an optional `None`-default parameter is now an override, not a
            second source of truth, so it is exempt
  TRF035 -- F401/F821/F822 are now accepted in modular files
  TRF041 -- guards (an `if` with no `else` that only raises or only logs) and a
            `DEFAULT_EXEMPT_ATTRIBUTES` list of framework plumbing are exempt

Dropped 39 allowlist entries that no longer fire, mostly mlinter#27 (the rule
engine now resolves a class's base chain across models instead of only inside
the linted file), which this branch had flagged upstream rather than worked
around:
  TRF018  16 -> 1   only `radio`, the one real case, is left
  TRF020   1 -> 0   the `axk2` false positive is gone
  TRF034  17 -> 5   the 12 cross-file false positives are gone; the 5 real
                    findings stay as follow-ups
  TRF029   2 -> 1   TRF035  5 -> 0   TRF041  14 -> 9
Each rule's comment block now describes what is left and why. Verified minimal:
emptying any of these six allowlists reproduces exactly the kept entries.

24 further entries (TRF002, TRF004, TRF005, TRF006, TRF009 `maskformer`,
TRF010, TRF019, TRF056) are also dead, but they were already dead before this
bump, so they are left alone here. TRF037's `x_clip` stays: the rule is opt-in,
so it never fires and the entry documents what would.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@tarekziade @claude
Regenerate modeling_rules.md for the synced rule prose
fcda31b 
Generated from utils/rules.toml by utils/check_modeling_rules_doc.py; picks up
the TRF029, TRF035 and TRF041 prose synced from transformers-mlinter main.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@tarekziade @claude
Drop 19 more allowlist entries that no longer fire
17f2903 
Swept every rule the way the earlier mlinter#27 pass was done: emptied each
allowlist, re-ran, and checked what actually fires. 213 entries -> 194.

"Does not fire" has four causes, so each candidate was classified rather than
deleted on the strength of a green run:
  - the model is compliant now
  - the pattern survives only in a generated file, copied from a parent
  - the model is grandfathered by cutoff_date, so it is never checked
  - the rule is opt-in and never runs at all
Only the first two are safe to drop. To separate them, each candidate rule was
run directly against the model's own source with grandfathering off.

Removed as compliant, verified rule-by-rule against the source (14):
  TRF005  9 -> 0   the rule accepts an explicit `_no_split_modules = None`, a
                   modular `AttributeError` sentinel, and raw-string elements
  TRF010  2 -> 0   both configs carry @strict(accept_kwargs=True)
  TRF006  4 -> 3   TRF009 15 -> 14 (maskformer)   TRF056 2 -> 1 (kimi_k25)

Removed as generated-only copies (5):
  TRF004  9 -> 4   hubert, sew, unispeech, unispeech_sat and wavlm define no
                   `tie_weights` in their modular files; the override exists
                   only in their generated modeling_*.py, copied from
                   Wav2Vec2ForCTC. wav2vec2 still fires, so the root cause stays
                   tracked -- listing all five implied five separate bugs.

Kept deliberately, and now documented in the file (5):
  TRF002  lighton_ocr -- `base_model_prefix = ""` is real and authored in its
          own modular file; the rule misses it only because it resolves base
          chains within a single file, so `LightOnOcrModel(Mistral3Model)` is
          not recognized as a pretrained class
  TRF019  blip_2, grounding_dino, omdet_turbo -- grandfathered by cutoff_date;
          they still set `_defaults` and fire once grandfathering is off
  TRF037  x_clip -- the rule is opt-in, so the entry records what would fire

Verified: emptying every allowlist now reproduces exactly the kept entries,
except those 5, each of which has a stated reason.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@tarekziade
Adopt the TRF034 trunk scoping from mlinter main
8e0e538 
mlinter#65 scoped TRF034 to the model's checkpointable token-mixing trunk. Sync
the rule prose and cut the allowlist from 5 models to 1.

`dinov3_convnext`, `openai` and `xcodec2` never set
`supports_gradient_checkpointing`, so `gradient_checkpointing_enable()` raises
for them rather than skipping a layer and there was nothing to report.
`tipsv2_dpt`'s three entries are DPT head layers, not a token-mixing stack.
`x_clip`'s `PromptGeneratorLayer` stays: an attention stack on a model that does
enable checkpointing, so it is a real finding and a follow-up for that model.

Over this checkout with every cutoff neutralised, TRF034 goes from 103 findings
over 76 models to 20 over 19, and every survivor is a transformer stack in a
checkpointable model.
@tarekziade

tarekziade commented Aug 28, 2026

Copy link
Copy Markdown
Collaborator Author

run-slow: bert, cohere_asr, convbert, dpr, esmfold2, fnet, gemma4_assistant, gemma4_unified_assistant, ovis2, qwen3_5, roberta, sam3, step3p7, vibevoice

@github-actions

github-actions Bot commented Aug 28, 2026

Copy link
Copy Markdown
Contributor

Workflow Run ⚙️

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

models: ["models/bert", "models/cohere_asr", "models/convbert", "models/dpr", "models/esmfold2", "models/fnet", "models/gemma4_assistant", "models/gemma4_unified_assistant", "models/ovis2", "models/qwen3_5", "models/roberta", "models/sam3", "models/step3p7", "models/vibevoice"]
quantizations: []

@github-actions

github-actions Bot commented Aug 28, 2026

Copy link
Copy Markdown
Contributor

CI Results

Workflow Run ⚙️

Commit Info

Context Commit Description
RUN 1d8e921c workflow commit (merge commit)
PR db03463c branch commit (from PR)
main 281dd533 base commit (on main)

✅ No failing test specific to this PR 🎉 👏 !

@github-actions

github-actions Bot commented Aug 28, 2026

Copy link
Copy Markdown
Contributor

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

run-slow: bert, cohere_asr, convbert, dpr, esmfold2, fnet, gemma4_assistant, gemma4_unified_assistant, ovis2, qwen3_5, roberta, sam3, step3p7, vibevoice

@tarekziade
tarekziade enabled auto-merge August 28, 2026 07:00
@github-actions

github-actions Bot commented Aug 28, 2026

Copy link
Copy Markdown
Contributor

CI recap

Dashboard: View test results in Grafana
Latest run: 33146857684:1
Result: failure | Jobs: 1 | Tests: 1,030 | Failures: 9 | Duration: 1h 42m

@tarekziade
tarekziade added this pull request to the merge queue Aug 28, 2026
@github-merge-queue
github-merge-queue Bot removed this pull request from the merge queue due to failed status checks Aug 28, 2026
@tarekziade
tarekziade added this pull request to the merge queue Aug 28, 2026
Merged via the queue into main with commit ba356d2 Aug 28, 2026
212 of 214 checks passed
@tarekziade
tarekziade deleted the mlinter-0.1.5 branch August 28, 2026 08:09
Qubitium added a commit to ModelCloud/OMY that referenced this pull request Aug 31, 2026
@Qubitium @ydshieh
@eryk-roch @claude @IlyasMoutawwakil @NanoCode012 @stevhliu @qgallouedec @remi-or @tarekziade @vasqu @sywangyi @glegendre01 @sharmax-vikas @Gronoxx Gustavo @Dovis01
Sync upstream transformers main into OMY (#14)
046f2d6 
* Retry get_daily_ci_runs on stale GitHub API cache (huggingface#48374)

* [debug] Poll GitHub API 10x to observe cache inconsistency in schedule runs

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Retry get_daily_ci_runs when GitHub API returns stale cache results

The GitHub Actions search index (used for event=/branch= filters) can lag
behind the database — different backend nodes return wildly different
total_count values (190, 238, 311, 413 observed for the same URL within
minutes), and the most-recent runs are missing from stale responses.

Detect staleness by checking whether the current GITHUB_RUN_ID appears in
the returned list; if absent, retry up to 3 times with a 30s delay.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [debug] Use get_daily_ci_runs in debug job; add current_run_id param

Add optional `current_run_id` parameter to `get_daily_ci_runs` so callers
can supply a known schedule run ID for stale-cache detection when
GITHUB_RUN_ID belongs to a non-schedule-triggered job (e.g. push).

Update the debug workflow job to call the Python function directly
(two quick back-to-back calls) instead of raw curl, so the retry logic
is exercised end-to-end and visible in the CI logs.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [debug] Force retry path with fake run ID to verify stale-detection logging

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Revert "[debug] Force retry path with fake run ID to verify stale-detection logging"

This reverts commit e49340f.

* [debug] Run get_daily_ci_runs 20 times to catch stale cache hit

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [debug] Sleep 60s between get_daily_ci_runs calls to hit different cache nodes

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Increase get_daily_ci_runs max_attempts from 3 to 5

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Add unit tests for get_daily_ci_runs stale-cache retry logic

- tests/utils/test_get_previous_daily_ci.py: 5 tests covering fresh hit
  (no retry), stale→fresh (one retry), all stale (max_attempts exhausted),
  no current_run_id (stale check skipped), and empty schedule fallback to
  workflow_run event
- utils/get_previous_daily_ci.py: ruff formatting fix (long print line)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Infer stale-check eligibility from current run's workflow_id

Instead of an explicit `current_run_id` argument, `get_daily_ci_runs` now
always fetches the current run's metadata (GITHUB_RUN_ID) upfront and
compares its `workflow_id` to the queried one:

- Same workflow → stale-cache detection applies (max_attempts=5, 30s sleep)
- Different workflow → skip (e.g. AMD CI querying Nvidia CI runs); left for
  a follow-up PR once the same-workflow case is confirmed stable

Also adds a TODO comment on the event=workflow_run fallback path (AMD CI)
and updates the 6 unit tests accordingly (new test for the different-workflow
case; all tests account for the upfront current-run lookup call).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Gate stale-cache check on GITHUB_EVENT_NAME == 'schedule'

Push- and dispatch-triggered runs won't appear in event=schedule results
even when the API is fresh, causing spurious retries. Only enable the
stale-cache retry loop when the current run is itself schedule-triggered.

Adds a corresponding unit test (test_non_schedule_event_skips_stale_check).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Revert debug changes to self-scheduled-caller.yml

Restore the workflow to its main-branch state; the debug polling job was
only needed to verify the stale-cache behaviour.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Condense get_daily_ci_runs comments and refer to huggingface#48374 for rationale

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* Fix `safe_open` mmap memory exhaustion on Windows by using `pread` backend (huggingface#48341)

* Fix `safe_open` mmap memory exhaustion on Windows by using `pread` backend

On Windows, memory-mapping safetensors files reserves copy-on-write commit
charge for the entire file. For large multi-shard checkpoints this exhausts
virtual memory. Switch to the `pread` backend on `win32` in both
`_load_pretrained_model` and `MtpModel` loading paths.

* Address review: add MPS backend fix to MtpModel, hoist backend selection before loop, use targeted sys.platform patch in test

* Address review: remove tests

* [ONNX] Skip affected models on torch 2.13 (two dynamo regressions) (huggingface#48191)

* [ONNX] Skip affected models on torch >= 2.13 (two dynamo ONNX regressions)

torch 2.13.0 introduced two regressions in dynamo ONNX export:
- pytorch/pytorch#194381: aten.sub type-promotion failure for scalar - int_tensor
- pytorch/pytorch#194382: aten.mul.Scalar missing ONNX decomposition

Skip all 35 affected model classes under `EXPORT_SKIPS["onnx"]` when
torch >= 2.13, using the existing skip infrastructure. Will be removed
once the upstream PyTorch fixes land.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Scope ONNX skip to torch == 2.13.x only (auto-runs on 2.14+)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Add missing BigBird subclasses to torch 2.13 ONNX skip list

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [ONNX] Clarify skip guard is torch == 2.13.x only (not >=)

Update comment and skip reason strings to say `torch == 2.13` instead of
`torch >= 2.13`, making it clear the guard auto-lifts on 2.14+.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [ONNX] Fix the torch 2.13 scalar regressions instead of skipping the models

Both regressions are a Python float meeting an integral tensor, whose promotion
torch 2.13 mishandles: `1.0 - int_mask` crashes the decomposition pass
(pytorch/pytorch#194381) and `int_mask * 2.0` reaches translation with no
registered ONNX decomposition (pytorch/pytorch#194382). On the same torch,
`1.0 - float_tensor` and `float_tensor * 2.0` export fine — so promoting the
tensor operand up front, to the dtype the op already produces, is enough. The
op and its overload are left alone, and the inserted cast carries the op's own
`meta` because for these elementwise cases it is the same value.

Decomposition also emits `mul.Scalar` with a *symbolic* second operand (a
division result, not a literal). There is no constant to promote there, so that
one is rewritten to `mul.Tensor`, which has the two-operand translation — the
same rewrite `_fix_remainder_scalar` makes for the same reason. Reachable
because the FX fixes run again right after `run_decompositions`.

This drops the 39 skip entries: the 15 affected families export again on 2.13,
including the vision models (Sam, SamHQ, GotOcr2, GroundingDino, SegGpt,
EfficientLoFTR, DeepseekOcr2).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Resolve the promoted-op set on first use, not at import

`exporter_onnx` is importable without torch — the CI job that imports transformers
with PIL only proved this the hard way — and naming `torch.ops.aten.*` overloads in
a module-level frozenset broke that with a `NameError` before anything ran.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* Say what the `mul.Scalar` operand is, and check it

Review raised two things. The `Node` check assumed a tensor operand: it is never one. Across the affected
families all 33 sites are an `operator.truediv` result, i.e. a `SymFloat`, which `mul.Tensor`'s translation
does take — so the rewrite stands, but the guard now names the forms that op accepts instead of trusting
the node type, and anything else keeps the `mul.Scalar` overload and fails visibly in translation.

The fixes are also deliberately not version-gated, which the docstring implied they were. Both rewrites
are semantics-preserving on any torch — a cast to the dtype the op already produces, and an overload swap
with the same meaning — so gating would only decide which torch exercises the path.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: IlyasMoutawwakil <moutawwakil.ilyas.tsi@gmail.com>
Co-authored-by: Ilyas Moutawwakil <57442720+IlyasMoutawwakil@users.noreply.github.com>

* [Qwen3VLMoe] Update `test_small_model_integration_test_batch` expected output (value drift) (huggingface#48376)

Update Qwen3VLMoe batch integration test expected output (value drift)

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* [LongcatFlash] Fix test_longcat_generation_cpu: use device_map="cpu" to avoid MoE disk offload issue (huggingface#48377)

* [LongcatFlash] Fix test_longcat_generation_cpu by using device_map="cpu"

`device_map="auto"` causes accelerate to offload MoE expert weights to disk,
which then fails to reload them due to an internal weight format incompatibility.
Since the test already requires large CPU RAM, use `device_map="cpu"` to keep
all weights in memory and avoid disk offloading entirely.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [LongcatFlash] Update golden string and skip test_longcat_generation_cpu on small runners

- `test_shortcat_generation`: update expected output to current model output (value drift)
- `test_longcat_generation_cpu`: replace `@require_large_cpu_ram` with
  `@require_torch_accelerator_memory(memory=1100)` — the 562B parameter model requires
  ~1,047 GiB of bfloat16 weights, far exceeding the CI runner budget (84 GiB single /
  168 GiB dual), and disk offloading fails due to MoE weight format incompatibility
  with accelerate

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* remove unused require_large_cpu_ram import

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* fix: flash-attn fallback failing on torch2.13 (huggingface#48388)

fix: probe the same flash-attn kernel version that the loader resolves

* [docs] Pass ContinuousBatchingConfig and sliding window models  (huggingface#48381)

docs

* Raise when a paged attention forward is called with no cache (huggingface#48297)

* Fall back to the base attention implementation when a paged attention forward has no cache

A standard forward on a model switched to a paged attention implementation (as done by
ContinuousBatchingManager) silently computed non-causal attention: sdpa_attention_paged_forward
treats cache as optional and mask creation early-exits to None for paged implementations.
Now, when the continuous batching kwargs are absent, the paged sdpa/flash forwards delegate to
their base implementation, the eager paged forward materializes a causal mask, and mask creation
resolves 2D padding masks with the base implementation's mask function.

* Raise instead of falling back when a paged attention forward has no cache

Drops the redirect to the base implementation and the paged-aware mask
creation, per review. `paged|sdpa` and `paged|eager` now refuse a call with
no paged cache rather than silently attending bidirectionally.

---------

Co-authored-by: Rémi Ouazan <83456801+remi-or@users.noreply.github.com>

* Bump transformers-mlinter to 0.1.5 and clear the new findings (huggingface#48259)

* future mlinter 0.1.5

* fix rule 58

* Add tokenizer tests for TRF038

Adds the missing tokenizer test files that mlinter's TRF038 reports, now that
the rule covers tokenization_*.py files.

* fix roberta

* Allowlist the three alias-only tokenizer modules for TRF038

bart, mobilebert and squeezebert started firing when mlinter 0.1.5 extended
TRF038 to tokenizer files. Their tokenization_*.py define no classes at all,
each being a two-line back-compat alias left by "rm slow tokenizers" (huggingface#40936),
which deleted their test files on purpose. A test file here could only assert
alias identity, and the aliased tokenizers are already covered by
tests/models/roberta and tests/models/bert.

Allowlisted as a placeholder: the better fix is upstream, where TRF038 should
skip a source file that defines no classes.

TRF038 now reports no findings.

* Route composite sub-configs through CONFIG_MAPPING for TRF009

mlinter 0.1.5 extended TRF009 from modeling files to every model file, which
surfaced six configs importing a sibling model's config class directly. Each now
resolves it through CONFIG_MAPPING with AutoConfig as the sub_configs marker,
which is what the rule prescribes and what 99 other composite configs already
do:

  cohere_asr               ParakeetEncoderConfig    -> parakeet_encoder
  esmfold2                 EsmcConfig               -> esmc
  gemma4_assistant         Gemma4TextConfig         -> gemma4_text
  gemma4_unified_assistant Gemma4UnifiedTextConfig  -> gemma4_unified_text
  ovis2                    Qwen2Config              -> qwen2
  sam3                     CLIPTextConfig           -> clip_text_model

cohere_asr already used CONFIG_MAPPING for instantiation, so only its
sub_configs value and import changed. Where the imported class also appeared in
a type annotation it is widened to PreTrainedConfig, and esmfold2's _init_nested
helper takes the class from CONFIG_MAPPING since it calls cls() directly.

Verified: default construction and to_dict/from_dict round-trip resolve the same
concrete config class as before for all six, and tests/models for the six models
pass (552 passed, 551 skipped).

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Backport mlinter rule-prose compression and allowlist the rest of TRF009

Two changes to utils/rules.toml.

Backports transformers-mlinter d0ac218 ("compress the rule explanation prose",
before: upstream's file as the base, with this repo's licence header and its
allowlist_models entries and their justification comments re-applied on top, so
the only remaining difference from the installed mlinter/rules.toml is the header
and those allowlists. make fix-repo regenerated
docs/source/en/modeling_rules.md.

Allowlists the ten TRF009 findings left after the CONFIG_MAPPING fixes, in three
groups, each with its reasoning recorded in the file:

- Back-compat tokenizer shims (bart, convbert, distilbert, fnet, mobilebert,
  squeezebert): the cross-model import is the module's whole purpose.
- generation_*.py (nemotron3_5_asr, nemotron_asr_streaming): the rule points at
  modular files, but the converter has no generation_*.py support, so the
  prescribed remedy does not exist yet.
- sam3: CLIPTextModelWithProjection is needed for checkpoint key compatibility.
  shieldgemma2: fixable with a modular file, left as follow-up.

make typing and make check-repo now pass.

* Fix vibevoice for mlinter 0.1.5 after rebase

VibeVoice (huggingface#40546) landed on main after this branch forked, so it was never
checked against the 0.1.5 rules and brought in two findings.

TRF058: VibeVoiceDiffusionHeadSinusoidalEmbedding created `freq` via
`self.register_buffer(...)`, which cannot be inherited and tweaked from a
modular file. Assign `nn.Buffer(..., persistent=False)` instead, which also
makes the `freq: torch.Tensor` linting workaround unnecessary. Same
non-persistent buffer and same forward output.

TRF009: allowlist `vibevoice`. generation_vibevoice.py imports a runtime cache
class from its own sub-model package; the family already routes its sub-configs
through CONFIG_MAPPING, and a cache class cannot come from CONFIG_MAPPING. It
also sits in a generation_*.py file, which the modular converter does not
support, the same blocker as the nemotron entries.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Adopt transformers-mlinter main: sync rule prose, drop dead allowlist entries

Re-ran the checker against transformers-mlinter git main (e6533da). Green, and
the bump makes a chunk of this branch's curation obsolete.

Synced the rule prose the repo copy mirrors, which had gone stale because the
described behavior changed upstream:
  TRF029 -- an optional `None`-default parameter is now an override, not a
            second source of truth, so it is exempt
  TRF035 -- F401/F821/F822 are now accepted in modular files
  TRF041 -- guards (an `if` with no `else` that only raises or only logs) and a
            `DEFAULT_EXEMPT_ATTRIBUTES` list of framework plumbing are exempt

Dropped 39 allowlist entries that no longer fire, mostly mlinter#27 (the rule
engine now resolves a class's base chain across models instead of only inside
the linted file), which this branch had flagged upstream rather than worked
around:
  TRF018  16 -> 1   only `radio`, the one real case, is left
  TRF020   1 -> 0   the `axk2` false positive is gone
  TRF034  17 -> 5   the 12 cross-file false positives are gone; the 5 real
                    findings stay as follow-ups
  TRF029   2 -> 1   TRF035  5 -> 0   TRF041  14 -> 9
Each rule's comment block now describes what is left and why. Verified minimal:
emptying any of these six allowlists reproduces exactly the kept entries.

24 further entries (TRF002, TRF004, TRF005, TRF006, TRF009 `maskformer`,
TRF010, TRF019, TRF056) are also dead, but they were already dead before this
bump, so they are left alone here. TRF037's `x_clip` stays: the rule is opt-in,
so it never fires and the entry documents what would.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Regenerate modeling_rules.md for the synced rule prose

Generated from utils/rules.toml by utils/check_modeling_rules_doc.py; picks up
the TRF029, TRF035 and TRF041 prose synced from transformers-mlinter main.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Drop 19 more allowlist entries that no longer fire

Swept every rule the way the earlier mlinter#27 pass was done: emptied each
allowlist, re-ran, and checked what actually fires. 213 entries -> 194.

"Does not fire" has four causes, so each candidate was classified rather than
deleted on the strength of a green run:
  - the model is compliant now
  - the pattern survives only in a generated file, copied from a parent
  - the model is grandfathered by cutoff_date, so it is never checked
  - the rule is opt-in and never runs at all
Only the first two are safe to drop. To separate them, each candidate rule was
run directly against the model's own source with grandfathering off.

Removed as compliant, verified rule-by-rule against the source (14):
  TRF005  9 -> 0   the rule accepts an explicit `_no_split_modules = None`, a
                   modular `AttributeError` sentinel, and raw-string elements
  TRF010  2 -> 0   both configs carry @strict(accept_kwargs=True)
  TRF006  4 -> 3   TRF009 15 -> 14 (maskformer)   TRF056 2 -> 1 (kimi_k25)

Removed as generated-only copies (5):
  TRF004  9 -> 4   hubert, sew, unispeech, unispeech_sat and wavlm define no
                   `tie_weights` in their modular files; the override exists
                   only in their generated modeling_*.py, copied from
                   Wav2Vec2ForCTC. wav2vec2 still fires, so the root cause stays
                   tracked -- listing all five implied five separate bugs.

Kept deliberately, and now documented in the file (5):
  TRF002  lighton_ocr -- `base_model_prefix = ""` is real and authored in its
          own modular file; the rule misses it only because it resolves base
          chains within a single file, so `LightOnOcrModel(Mistral3Model)` is
          not recognized as a pretrained class
  TRF019  blip_2, grounding_dino, omdet_turbo -- grandfathered by cutoff_date;
          they still set `_defaults` and fire once grandfathering is off
  TRF037  x_clip -- the rule is opt-in, so the entry records what would fire

Verified: emptying every allowlist now reproduces exactly the kept entries,
except those 5, each of which has a stated reason.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Adopt the TRF034 trunk scoping from mlinter main

mlinter#65 scoped TRF034 to the model's checkpointable token-mixing trunk. Sync
the rule prose and cut the allowlist from 5 models to 1.

`dinov3_convnext`, `openai` and `xcodec2` never set
`supports_gradient_checkpointing`, so `gradient_checkpointing_enable()` raises
for them rather than skipping a layer and there was nothing to report.
`tipsv2_dpt`'s three entries are DPT head layers, not a token-mixing stack.
`x_clip`'s `PromptGeneratorLayer` stays: an attention stack on a model that does
enable checkpointing, so it is a real finding and a follow-up for that model.

Over this checkout with every cutoff neutralised, TRF034 goes from 103 findings
over 76 models to 20 over 19, and every survivor is a transformer stack in a
checkpointable model.

* Pin transformers-mlinter 0.1.5

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

* [CB] Fix wrong device scoping (huggingface#48370)

* Add device scoping

* Added a regression test

* [Improvement] Make gated delta rule more explicit  (huggingface#47625)

* MAke gdr more explicit and support per-channel decay

* Nits and tests

* regenerated

* Delete the KDA mode of GDN

* Bake in the UT transform

* Review compliance 1/n

* Apply suggestions from code review

Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>

* Readability

* Review 2/n

* Memory and speed optims

* Added test file

* Fix export

* Merge mess up fix

* fix2

* Fix in place identity

* Review compliance

* second review 1/n

* second review 2/n

* second review 3/n

* second review 4/n

* second review 5/n

* Fix repo

* Last review

* Remove test class

* Make fix repo

* Fix BPE tokenizer test

---------

Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>

* fix some failure in xpu (huggingface#48252)

* fix some failure in xpu

Signed-off-by: Wang, Yi <yi.a.wang@intel.com>

* fmt

Signed-off-by: Wang, Yi <yi.a.wang@intel.com>

* update

Signed-off-by: Wang, Yi <yi.a.wang@intel.com>

* update

Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>

---------

Signed-off-by: Wang, Yi <yi.a.wang@intel.com>
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>

* Update Tailscale action version in workflow (huggingface#48394)

* skip mtp slow tests for now (huggingface#48328) (huggingface#48329)

skip mtp slow tests for now

* [VibeVoice] Skip generate export tests (flaky) (huggingface#48396)

* [VibeVoice] Skip generate export tests (flaky, decompose_prefill_decode incompatible)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* [VibeVoice] Skip flaky static-cache compile tests

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* Fix interval merge invariant in _find_disjoint (huggingface#47860)

Fix interval merge invariant in `_find_disjoint`

`_find_disjoint` implements a sweep-line merge over tensor memory ranges,
but assigns `last_stop = stop` instead of tracking the running maximum.
When a tensor is strictly contained inside an earlier, larger one,
`last_stop` shrinks to the contained tensor's end pointer, so a subsequent
tensor that genuinely overlaps the larger one is classified as disjoint.

Use `last_stop = max(last_stop, stop)`, the standard sweep-line invariant.

Note on impact: this is a latent correctness fix, not a user-facing bug fix.
Triggering the mis-classification requires strict containment, which in turn
guarantees the resulting group is not "identical" in `_find_identical`, so
`remove_tied_weights_from_state_dict` raises the same RuntimeError either way.
Verified exhaustively over all 3- and 4-interval configurations (54k cases):
the corrected version matches ground-truth connected components in 100% of
cases, and the raise-vs-succeed outcome never differs. The observable
improvement is that the error message now reports the complete set of
overlapping tensors instead of a truncated one.

* Fix incorrect tuple return annotations on forward methods returning a Tensor (huggingface#48359)

A number of `forward` methods declare a `tuple[...]` return annotation but
return a bare `torch.Tensor`. Same class of issue as huggingface#45208, fixed for
`Qwen3MoeSparseMoeBlock` in huggingface#45352.

Each of the 29 classes touched here was instantiated and called with a minimal
config, and the returned object's type observed to be `torch.Tensor`. Annotation
only: no runtime behaviour is affected.

Eleven annotations were edited in `modular_*.py` sources and the corresponding
`modeling_*.py` files regenerated with `utils/modular_model_converter.py`;
`utils/check_modular_conversion.py` and `utils/check_copies.py` both pass.

Co-authored-by: Gustavo <gustavo.vp350@gmail.com>

* Avoid print to stdout that fails the job `check_failed_tests` job (huggingface#48391)

fix

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* [Docs]: Update GLM 5.3 (huggingface#48401)

* docs: update GLM 5.3

Signed-off-by: Shijin Zhang <75300765+Dovis01@users.noreply.github.com>

* docs: update toc

Signed-off-by: Shijin Zhang <75300765+Dovis01@users.noreply.github.com>

* fix

Signed-off-by: Shijin Zhang <75300765+Dovis01@users.noreply.github.com>

---------

Signed-off-by: Shijin Zhang <75300765+Dovis01@users.noreply.github.com>

---------

Signed-off-by: Wang, Yi <yi.a.wang@intel.com>
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>
Signed-off-by: Shijin Zhang <75300765+Dovis01@users.noreply.github.com>
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
Co-authored-by: eryk-roch <eryk.roch.rybak@intel.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: IlyasMoutawwakil <moutawwakil.ilyas.tsi@gmail.com>
Co-authored-by: Ilyas Moutawwakil <57442720+IlyasMoutawwakil@users.noreply.github.com>
Co-authored-by: NanoCode012 <nano@axolotl.ai>
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>
Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com>
Co-authored-by: Rémi Ouazan <83456801+remi-or@users.noreply.github.com>
Co-authored-by: Tarek Ziade <tarek@ziade.org>
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>
Co-authored-by: Wang, Yi <yi.a.wang@intel.com>
Co-authored-by: Guillaume LEGENDRE <glegendre01@gmail.com>
Co-authored-by: Vikas Sharma <vikasharmadev8@gmail.com>
Co-authored-by: Gustavo Dias Apolinário <100954782+Gronoxx@users.noreply.github.com>
Co-authored-by: Gustavo <gustavo.vp350@gmail.com>
Co-authored-by: Shijin Zhang <75300765+Dovis01@users.noreply.github.com>
sbucaille pushed a commit to sbucaille/transformers that referenced this pull request Sep 16, 2026
@tarekziade @claude @sbucaille
Bump transformers-mlinter to 0.1.5 and clear the new findings (huggin…
3556285 
…gface#48259)

* future mlinter 0.1.5

* fix rule 58

* Add tokenizer tests for TRF038

Adds the missing tokenizer test files that mlinter's TRF038 reports, now that
the rule covers tokenization_*.py files.

* fix roberta

* Allowlist the three alias-only tokenizer modules for TRF038

bart, mobilebert and squeezebert started firing when mlinter 0.1.5 extended
TRF038 to tokenizer files. Their tokenization_*.py define no classes at all,
each being a two-line back-compat alias left by "rm slow tokenizers" (huggingface#40936),
which deleted their test files on purpose. A test file here could only assert
alias identity, and the aliased tokenizers are already covered by
tests/models/roberta and tests/models/bert.

Allowlisted as a placeholder: the better fix is upstream, where TRF038 should
skip a source file that defines no classes.

TRF038 now reports no findings.

* Route composite sub-configs through CONFIG_MAPPING for TRF009

mlinter 0.1.5 extended TRF009 from modeling files to every model file, which
surfaced six configs importing a sibling model's config class directly. Each now
resolves it through CONFIG_MAPPING with AutoConfig as the sub_configs marker,
which is what the rule prescribes and what 99 other composite configs already
do:

  cohere_asr               ParakeetEncoderConfig    -> parakeet_encoder
  esmfold2                 EsmcConfig               -> esmc
  gemma4_assistant         Gemma4TextConfig         -> gemma4_text
  gemma4_unified_assistant Gemma4UnifiedTextConfig  -> gemma4_unified_text
  ovis2                    Qwen2Config              -> qwen2
  sam3                     CLIPTextConfig           -> clip_text_model

cohere_asr already used CONFIG_MAPPING for instantiation, so only its
sub_configs value and import changed. Where the imported class also appeared in
a type annotation it is widened to PreTrainedConfig, and esmfold2's _init_nested
helper takes the class from CONFIG_MAPPING since it calls cls() directly.

Verified: default construction and to_dict/from_dict round-trip resolve the same
concrete config class as before for all six, and tests/models for the six models
pass (552 passed, 551 skipped).

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Backport mlinter rule-prose compression and allowlist the rest of TRF009

Two changes to utils/rules.toml.

Backports transformers-mlinter d0ac218 ("compress the rule explanation prose",
before: upstream's file as the base, with this repo's licence header and its
allowlist_models entries and their justification comments re-applied on top, so
the only remaining difference from the installed mlinter/rules.toml is the header
and those allowlists. make fix-repo regenerated
docs/source/en/modeling_rules.md.

Allowlists the ten TRF009 findings left after the CONFIG_MAPPING fixes, in three
groups, each with its reasoning recorded in the file:

- Back-compat tokenizer shims (bart, convbert, distilbert, fnet, mobilebert,
  squeezebert): the cross-model import is the module's whole purpose.
- generation_*.py (nemotron3_5_asr, nemotron_asr_streaming): the rule points at
  modular files, but the converter has no generation_*.py support, so the
  prescribed remedy does not exist yet.
- sam3: CLIPTextModelWithProjection is needed for checkpoint key compatibility.
  shieldgemma2: fixable with a modular file, left as follow-up.

make typing and make check-repo now pass.

* Fix vibevoice for mlinter 0.1.5 after rebase

VibeVoice (huggingface#40546) landed on main after this branch forked, so it was never
checked against the 0.1.5 rules and brought in two findings.

TRF058: VibeVoiceDiffusionHeadSinusoidalEmbedding created `freq` via
`self.register_buffer(...)`, which cannot be inherited and tweaked from a
modular file. Assign `nn.Buffer(..., persistent=False)` instead, which also
makes the `freq: torch.Tensor` linting workaround unnecessary. Same
non-persistent buffer and same forward output.

TRF009: allowlist `vibevoice`. generation_vibevoice.py imports a runtime cache
class from its own sub-model package; the family already routes its sub-configs
through CONFIG_MAPPING, and a cache class cannot come from CONFIG_MAPPING. It
also sits in a generation_*.py file, which the modular converter does not
support, the same blocker as the nemotron entries.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Adopt transformers-mlinter main: sync rule prose, drop dead allowlist entries

Re-ran the checker against transformers-mlinter git main (e6533da). Green, and
the bump makes a chunk of this branch's curation obsolete.

Synced the rule prose the repo copy mirrors, which had gone stale because the
described behavior changed upstream:
  TRF029 -- an optional `None`-default parameter is now an override, not a
            second source of truth, so it is exempt
  TRF035 -- F401/F821/F822 are now accepted in modular files
  TRF041 -- guards (an `if` with no `else` that only raises or only logs) and a
            `DEFAULT_EXEMPT_ATTRIBUTES` list of framework plumbing are exempt

Dropped 39 allowlist entries that no longer fire, mostly mlinter#27 (the rule
engine now resolves a class's base chain across models instead of only inside
the linted file), which this branch had flagged upstream rather than worked
around:
  TRF018  16 -> 1   only `radio`, the one real case, is left
  TRF020   1 -> 0   the `axk2` false positive is gone
  TRF034  17 -> 5   the 12 cross-file false positives are gone; the 5 real
                    findings stay as follow-ups
  TRF029   2 -> 1   TRF035  5 -> 0   TRF041  14 -> 9
Each rule's comment block now describes what is left and why. Verified minimal:
emptying any of these six allowlists reproduces exactly the kept entries.

24 further entries (TRF002, TRF004, TRF005, TRF006, TRF009 `maskformer`,
TRF010, TRF019, TRF056) are also dead, but they were already dead before this
bump, so they are left alone here. TRF037's `x_clip` stays: the rule is opt-in,
so it never fires and the entry documents what would.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Regenerate modeling_rules.md for the synced rule prose

Generated from utils/rules.toml by utils/check_modeling_rules_doc.py; picks up
the TRF029, TRF035 and TRF041 prose synced from transformers-mlinter main.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Drop 19 more allowlist entries that no longer fire

Swept every rule the way the earlier mlinter#27 pass was done: emptied each
allowlist, re-ran, and checked what actually fires. 213 entries -> 194.

"Does not fire" has four causes, so each candidate was classified rather than
deleted on the strength of a green run:
  - the model is compliant now
  - the pattern survives only in a generated file, copied from a parent
  - the model is grandfathered by cutoff_date, so it is never checked
  - the rule is opt-in and never runs at all
Only the first two are safe to drop. To separate them, each candidate rule was
run directly against the model's own source with grandfathering off.

Removed as compliant, verified rule-by-rule against the source (14):
  TRF005  9 -> 0   the rule accepts an explicit `_no_split_modules = None`, a
                   modular `AttributeError` sentinel, and raw-string elements
  TRF010  2 -> 0   both configs carry @strict(accept_kwargs=True)
  TRF006  4 -> 3   TRF009 15 -> 14 (maskformer)   TRF056 2 -> 1 (kimi_k25)

Removed as generated-only copies (5):
  TRF004  9 -> 4   hubert, sew, unispeech, unispeech_sat and wavlm define no
                   `tie_weights` in their modular files; the override exists
                   only in their generated modeling_*.py, copied from
                   Wav2Vec2ForCTC. wav2vec2 still fires, so the root cause stays
                   tracked -- listing all five implied five separate bugs.

Kept deliberately, and now documented in the file (5):
  TRF002  lighton_ocr -- `base_model_prefix = ""` is real and authored in its
          own modular file; the rule misses it only because it resolves base
          chains within a single file, so `LightOnOcrModel(Mistral3Model)` is
          not recognized as a pretrained class
  TRF019  blip_2, grounding_dino, omdet_turbo -- grandfathered by cutoff_date;
          they still set `_defaults` and fire once grandfathering is off
  TRF037  x_clip -- the rule is opt-in, so the entry records what would fire

Verified: emptying every allowlist now reproduces exactly the kept entries,
except those 5, each of which has a stated reason.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* Adopt the TRF034 trunk scoping from mlinter main

mlinter#65 scoped TRF034 to the model's checkpointable token-mixing trunk. Sync
the rule prose and cut the allowlist from 5 models to 1.

`dinov3_convnext`, `openai` and `xcodec2` never set
`supports_gradient_checkpointing`, so `gradient_checkpointing_enable()` raises
for them rather than skipping a layer and there was nothing to report.
`tipsv2_dpt`'s three entries are DPT head layers, not a token-mixing stack.
`x_clip`'s `PromptGeneratorLayer` stays: an attention stack on a model that does
enable checkpointing, so it is a real finding and a follow-up for that model.

Over this checkout with every cutoff neutralised, TRF034 goes from 103 findings
over 76 models to 20 over 19, and every survivor is a transformer stack in a
checkpointable model.

* Pin transformers-mlinter 0.1.5

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@github-actions github-actions[bot] github-actions[bot] left review comments
@ArthurZucker ArthurZucker ArthurZucker approved these changes
@zucchini-nlp zucchini-nlp zucchini-nlp approved these changes
@Cyrilvallez Cyrilvallez Awaiting requested review from Cyrilvallez
@molbap molbap Awaiting requested review from molbap
@vasqu vasqu Awaiting requested review from vasqu

Assignees

@tarekziade tarekziade

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@tarekziade @HuggingFaceDocBuilderDev @ArthurZucker @zucchini-nlp