Skip models that cannot currently init on Transformers v5 (#28471)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
(cherry picked from commit 51c599f0ec)

.buildkite/test-pipeline.yaml

@@ -868,12 +868,12 @@ steps:
   optional: true
   commands:
     - pip install --upgrade git+https://github.com/huggingface/transformers
-    - pytest -v -s tests/models/test_initialization.py
+    - pytest -v -s tests/models/test_initialization.py -k 'not (Gemma3 or ModernBert or Qwen2_5_VL or Qwen2_5vl or Qwen2VL or TransformersMultiModalEmbeddingModel or TransformersMultiModalForSequenceClassification or Ultravox or Phi4Multimodal or LlavaNextVideo or MiniCPMO or Lfm2Moe or PaliGemma or RobertaForSequenceClassification or Ovis2_5 or Fuyu or DeepseekOCR or KimiVL)'
     - pytest -v -s tests/models/test_transformers.py
-    - pytest -v -s tests/models/multimodal/processing/
+    # - pytest -v -s tests/models/multimodal/processing/
-    - pytest -v -s tests/models/multimodal/test_mapping.py
+    - pytest -v -s tests/models/multimodal/test_mapping.py -k 'not (Gemma3 or Qwen2VL or Qwen2_5_VL)'
     - python3 examples/offline_inference/basic/chat.py
-    - python3 examples/offline_inference/vision_language.py --model-type qwen2_5_vl
+    # - python3 examples/offline_inference/vision_language.py --model-type qwen2_5_vl
     # Whisper needs spawn method to avoid deadlock
     - VLLM_WORKER_MULTIPROC_METHOD=spawn python3 examples/offline_inference/audio_language.py --model-type whisper
 

vllm/model_executor/model_loader/weight_utils.py

@@ -82,7 +82,8 @@ enable_hf_transfer()
 
 class DisabledTqdm(tqdm):
     def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs, disable=True)
+        kwargs["disable"] = True
+        super().__init__(*args, **kwargs)
 
 
 def get_lock(model_name_or_path: str | Path, cache_dir: str | None = None):

vllm/model_executor/models/whisper.py

@@ -13,7 +13,6 @@ from transformers import (
     BatchFeature,
     WhisperConfig,
     WhisperFeatureExtractor,
-    WhisperProcessor,
 )
 from transformers.models.whisper.modeling_whisper import sinusoids
 
@@ -660,16 +659,6 @@ class WhisperProcessingInfo(BaseProcessingInfo):
     def get_hf_config(self) -> WhisperConfig:
         return self.ctx.get_hf_config(WhisperConfig)
 
-    def get_hf_processor(self, **kwargs: object) -> WhisperProcessor:
-        # HACK: Transformers 4.53.2 has issue with whisper tokenizer to
-        # initialize processor. We use a monkeypatch to fix it here.
-        # See: https://github.com/vllm-project/vllm/issues/20224
-        processor_class = WhisperProcessor
-        tokenizer_class = ("WhisperTokenizer", "WhisperTokenizerFast")
-        if processor_class.tokenizer_class != tokenizer_class:
-            processor_class.tokenizer_class = tokenizer_class
-        return self.ctx.get_hf_processor(processor_class, **kwargs)
-
     def get_supported_mm_limits(self) -> Mapping[str, int | None]:
         return {"audio": 1}