From 7e51939e25c34e6527a4971345c95dbefb029fc9 Mon Sep 17 00:00:00 2001 From: Andreas Karatzas Date: Fri, 24 Jul 2026 14:33:53 -0500 Subject: [PATCH] [CI] Avoid unnecessary Hugging Face metadata requests (#49508) Signed-off-by: Andreas Karatzas --- vllm/config/model.py | 9 ++++++--- vllm/model_executor/model_loader/default_loader.py | 2 +- 2 files changed, 7 insertions(+), 4 deletions(-) diff --git a/vllm/config/model.py b/vllm/config/model.py index 0e7621c2b1..1f62351c52 100644 --- a/vllm/config/model.py +++ b/vllm/config/model.py @@ -579,9 +579,9 @@ class ModelConfig: self.hf_text_config, "attention_chunk_size", None ) self.encoder_config = self._get_encoder_config() - self.hf_image_processor_config = get_hf_image_processor_config( - self.model, hf_token=self.hf_token, revision=self.revision - ) + # Image-processor metadata is only consumed by multimodal models. + # Probing it for text-only models causes avoidable Hub requests. + self.hf_image_processor_config: dict[str, Any] = {} architectures = self.architectures registry = self.registry @@ -703,6 +703,9 @@ class ModelConfig: # Init multimodal config if needed if self._model_info.supports_multimodal: + self.hf_image_processor_config = get_hf_image_processor_config( + self.model, hf_token=self.hf_token, revision=self.revision + ) if ( mm_encoder_tp_mode == "data" and not self._model_info.supports_multimodal_encoder_tp_data diff --git a/vllm/model_executor/model_loader/default_loader.py b/vllm/model_executor/model_loader/default_loader.py index 3ea76f4d9b..039a577f80 100644 --- a/vllm/model_executor/model_loader/default_loader.py +++ b/vllm/model_executor/model_loader/default_loader.py @@ -220,7 +220,7 @@ class DefaultModelLoader(BaseModelLoader): # safetensors file. Using both breaks. # Here, we download the `model.safetensors.index.json` and filter # any files not found in the index. - if not is_local: + if not is_local and len(hf_weights_files) > 1: download_safetensors_index_file_from_hf( model_name_or_path, index_file,