[Misc] Clean up processor tests (#14771)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

tests/models/multimodal/processing/test_h2ovl.py

@@ -96,14 +96,14 @@ def _run_check(
     tokenizer = processor.info.get_tokenizer()
     config = processor.info.get_hf_config()
 
+    prompt = "<image>" * len(images)
     mm_data = {"image": images}
 
     total_expected_num_patches = sum(
         _get_expected_num_patches(config, image, len(images), min_num, max_num)
         for image in images)
 
-    processed_inputs = processor.apply("<image>" * len(images), mm_data,
-                                       mm_processor_kwargs)
+    processed_inputs = processor.apply(prompt, mm_data, mm_processor_kwargs)
 
     # Ensure we have the right number of placeholders per num_crops size
     image_token_id = tokenizer.convert_tokens_to_ids("<IMG_CONTEXT>")
@@ -152,9 +152,7 @@ def test_processor_override(
     }
 
     ctx = build_model_context(
-        model_name=model_id,
-        tokenizer_name=model_id,
-        trust_remote_code=True,
+        model_id,
         mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
         limit_mm_per_prompt={"image": len(size_factors)},
     )

tests/models/multimodal/processing/test_idefics3.py

@@ -9,10 +9,8 @@ from vllm.transformers_utils.tokenizer import cached_tokenizer_from_config
 from ....conftest import _ImageAssets
 from ...utils import build_model_context
 
-models = ["HuggingFaceM4/Idefics3-8B-Llama3"]
 
-
-@pytest.mark.parametrize("model", models)
+@pytest.mark.parametrize("model_id", ["HuggingFaceM4/Idefics3-8B-Llama3"])
 # yapf: disable
 @pytest.mark.parametrize(
     ("mm_processor_kwargs", "expected_toks_per_img"),
@@ -25,7 +23,7 @@ models = ["HuggingFaceM4/Idefics3-8B-Llama3"]
 @pytest.mark.parametrize("kwargs_on_init", [True, False])
 def test_processor_override(
     image_assets: _ImageAssets,
-    model: str,
+    model_id: str,
     mm_processor_kwargs: dict[str, object],
     expected_toks_per_img: int,
     num_imgs: int,
@@ -36,9 +34,7 @@ def test_processor_override(
     # in this test and assume that the kwargs will be correctly expanded by
     # the partial when calling the custom input processor.
     ctx = build_model_context(
-        model_name=model,
-        tokenizer_name=model,
-        trust_remote_code=True,
+        model_id,
         mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
         limit_mm_per_prompt={"image": num_imgs},
     )

tests/models/multimodal/processing/test_internvl.py

@@ -56,14 +56,14 @@ def _run_check(
     tokenizer = processor.info.get_tokenizer()
     config = processor.info.get_hf_config()
 
+    prompt = "<image>" * len(images)
     mm_data = {"image": images}
 
     total_expected_num_patches = sum(
         _get_expected_num_patches(config, image, len(images), min_num, max_num)
         for image in images)
 
-    processed_inputs = processor.apply("<image>" * len(images), mm_data,
-                                       mm_processor_kwargs)
+    processed_inputs = processor.apply(prompt, mm_data, mm_processor_kwargs)
 
     # Ensure we have the right number of placeholders per num_crops size
     image_token_id = tokenizer.convert_tokens_to_ids("<IMG_CONTEXT>")
@@ -109,9 +109,7 @@ def test_processor_override(
     }
 
     ctx = build_model_context(
-        model_name=model_id,
-        tokenizer_name=model_id,
-        trust_remote_code=True,
+        model_id,
         mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
         limit_mm_per_prompt={"image": len(size_factors)},
     )

tests/models/multimodal/processing/test_llava_next.py

@@ -36,8 +36,7 @@ def _validate_image_max_tokens_one(
 @pytest.mark.parametrize("model_id", ["llava-hf/llava-v1.6-mistral-7b-hf"])
 def test_processor_max_tokens(model_id):
     ctx = build_model_context(
-        model_name=model_id,
-        tokenizer_name=model_id,
+        model_id,
         mm_processor_kwargs=None,
         limit_mm_per_prompt={"image": 1},
     )
@@ -136,8 +135,7 @@ def _test_image_prompt_replacements(
 @pytest.mark.parametrize("num_imgs", [1, 2])
 def test_processor_prompt_replacements_regression(model_id, num_imgs):
     ctx = build_model_context(
-        model_name=model_id,
-        tokenizer_name=model_id,
+        model_id,
         mm_processor_kwargs=None,
         limit_mm_per_prompt={"image": num_imgs},
     )
@@ -166,8 +164,7 @@ def test_processor_prompt_replacements_regression(model_id, num_imgs):
 @pytest.mark.parametrize("num_imgs", [1])
 def test_processor_prompt_replacements_all(model_id, num_imgs):
     ctx = build_model_context(
-        model_name=model_id,
-        tokenizer_name=model_id,
+        model_id,
         mm_processor_kwargs=None,
         limit_mm_per_prompt={"image": num_imgs},
     )

tests/models/multimodal/processing/test_llava_onevision.py

@@ -37,8 +37,7 @@ def _validate_image_max_tokens_one(
                          ["llava-hf/llava-onevision-qwen2-0.5b-ov-hf"])
 def test_processor_max_tokens(model_id):
     ctx = build_model_context(
-        model_name=model_id,
-        tokenizer_name=model_id,
+        model_id,
         mm_processor_kwargs=None,
         limit_mm_per_prompt={"image": 1},
     )
@@ -136,8 +135,7 @@ def _test_image_prompt_replacements(
 @pytest.mark.parametrize("num_imgs", [1, 2])
 def test_processor_prompt_replacements_regression(model_id, num_imgs):
     ctx = build_model_context(
-        model_name=model_id,
-        tokenizer_name=model_id,
+        model_id,
         mm_processor_kwargs=None,
         limit_mm_per_prompt={"image": num_imgs},
     )
@@ -167,8 +165,7 @@ def test_processor_prompt_replacements_regression(model_id, num_imgs):
 @pytest.mark.parametrize("num_imgs", [1])
 def test_processor_prompt_replacements_all(model_id, num_imgs):
     ctx = build_model_context(
-        model_name=model_id,
-        tokenizer_name=model_id,
+        model_id,
         mm_processor_kwargs=None,
         limit_mm_per_prompt={"image": num_imgs},
     )

tests/models/multimodal/processing/test_phi3v.py

@@ -35,9 +35,7 @@ def test_processor_override(
     from vllm.model_executor.models.phi3v import _IMAGE_TOKEN_ID
 
     ctx = build_model_context(
-        model_name=model_id,
-        tokenizer_name=model_id,
-        trust_remote_code=True,
+        model_id,
         mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
         limit_mm_per_prompt={"image": num_imgs},
     )

tests/models/multimodal/processing/test_qwen2_vl.py

@@ -30,8 +30,7 @@ def test_processor_override(
 ):
     """Ensure Qwen2VLMultiModalProcessor handles min/max pixels properly."""
     ctx = build_model_context(
-        model_name=model_id,
-        tokenizer_name=model_id,
+        model_id,
         mm_processor_kwargs=mm_processor_kwargs if kwargs_on_init else None,
         limit_mm_per_prompt={"image": num_imgs},
     )

tests/models/utils.py

@@ -10,6 +10,8 @@ from vllm.config import ModelConfig, TaskOption
 from vllm.inputs import InputContext
 from vllm.sequence import Logprob, PromptLogprobs, SampleLogprobs
 
+from .registry import HF_EXAMPLE_MODELS
+
 TokensText = tuple[list[int], str]
 
 
@@ -250,10 +252,8 @@ def check_logprobs_close(
 
 
 def build_model_context(
-    model_name: str,
+    model_id: str,
     task: TaskOption = "auto",
-    tokenizer_name: Optional[str] = None,
-    trust_remote_code: bool = False,
     dtype: Optional[Union[str, torch.dtype]] = None,
     mm_processor_kwargs: Optional[dict] = None,
     limit_mm_per_prompt: Optional[dict] = None,
@@ -262,9 +262,7 @@ def build_model_context(
     """Creates an InputContext for a given model.
 
     Args:
-        model_name: Name of the model being considered.
-        tokenizer_name: Name of the tokenizer being considered.
-        trust_remote_code: Whether or not to allow loading remote code.
+        model_id: ID of the model being considered.
         mm_processor_kwargs: optional processor kwargs for to be leveraged
             in the input processor, mapper, dummy data creation, etc.
         limit_mm_per_prompt: Multimodal limits.
@@ -272,21 +270,24 @@ def build_model_context(
     Returns:
         InputContext for the model being considered.
     """
-    if tokenizer_name is None:
-        tokenizer_name = model_name
+    model_info = HF_EXAMPLE_MODELS.find_hf_info(model_id)
+    model_info.check_available_online(on_fail="skip")
+    model_info.check_transformers_version(on_fail="skip")
+
     if dtype is None:
         dtype = "half"
 
     model_config = ModelConfig(
-        model_name,
+        model_id,
         task=task,
-        tokenizer=tokenizer_name,
-        tokenizer_mode="auto",
-        trust_remote_code=trust_remote_code,
+        tokenizer=model_info.tokenizer or model_id,
+        tokenizer_mode=model_info.tokenizer_mode,
+        trust_remote_code=model_info.trust_remote_code,
         dtype=dtype,
         seed=0,
         mm_processor_kwargs=mm_processor_kwargs,
         limit_mm_per_prompt=limit_mm_per_prompt,
         disable_mm_preprocessor_cache=disable_mm_preprocessor_cache,
+        hf_overrides=model_info.hf_overrides,
     )
     return InputContext(model_config)