[tests] fix tests

tests/models/multimodal/processing/test_llava_onevision.py

@@ -3,6 +3,7 @@
 import itertools
 from functools import partial
 
+import os
 import pytest
 from PIL import Image
 from pqdm.threads import pqdm
@@ -12,6 +13,7 @@ from vllm.multimodal.parse import ImageSize
 from vllm.multimodal.processing import BaseMultiModalProcessor
 
 from ...utils import build_model_context
+from ....utils import models_path_prefix
 
 
 def _validate_image_max_tokens_one(
@@ -33,7 +35,7 @@ def _validate_image_max_tokens_one(
 @pytest.mark.skip("This test takes around 5 minutes to run. "
                   "Comment this out to run it manually.")
 @pytest.mark.parametrize("model_id",
-                         ["llava-hf/llava-onevision-qwen2-0.5b-ov-hf"])
+                         [os.path.join(models_path_prefix, "llava-hf/llava-onevision-qwen2-0.5b-ov-hf")])
 def test_processor_max_tokens(model_id):
     ctx = build_model_context(
         model_id,
@@ -127,7 +129,7 @@ def _test_image_prompt_replacements(
 
 
 @pytest.mark.parametrize("model_id",
-                         ["llava-hf/llava-onevision-qwen2-0.5b-ov-hf"])
+                         [os.path.join(models_path_prefix, "llava-hf/llava-onevision-qwen2-0.5b-ov-hf")])
 @pytest.mark.parametrize("num_imgs", [1, 2])
 def test_processor_prompt_replacements_regression(model_id, num_imgs):
     ctx = build_model_context(

tests/models/multimodal/processing/test_mllama.py

 # SPDX-License-Identifier: Apache-2.0
 """Tests for mllama's multimodal preprocessing and profiling."""
+import os
 import pytest
 from transformers import MllamaConfig
 
@@ -7,10 +8,11 @@ from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.profiling import MultiModalProfiler
 
 from ...utils import build_model_context
+from ....utils import models_path_prefix
 
 
 @pytest.mark.parametrize("model_id",
-                         ["meta-llama/Llama-3.2-11B-Vision-Instruct"])
+                         [os.path.join(models_path_prefix, "meta-llama/Llama-3.2-11B-Vision-Instruct")])
 @pytest.mark.parametrize("max_model_len", [4096, 8192, 25600, 131072])
 @pytest.mark.parametrize("max_num_seqs", [1, 2, 8])
 def test_profiling(

tests/models/multimodal/processing/test_phi3v.py

 # SPDX-License-Identifier: Apache-2.0
 """Tests for phi3v's multimodal preprocessing kwargs."""
+import os
 import pytest
 
 from vllm.multimodal import MULTIMODAL_REGISTRY
 
 from ....conftest import _ImageAssets
 from ...utils import build_model_context
+from ....utils import models_path_prefix
 
 
-@pytest.mark.parametrize("model_id", ["microsoft/Phi-3.5-vision-instruct"])
+@pytest.mark.parametrize("model_id", [os.path.join(models_path_prefix, "microsoft/Phi-3.5-vision-instruct")])
 # yapf: disable
 @pytest.mark.parametrize(
     ("mm_processor_kwargs", "expected_toks_per_img"),

tests/models/multimodal/processing/test_phi4mm.py

 # SPDX-License-Identifier: Apache-2.0
 """Tests for phi4mm's multimodal preprocessing kwargs."""
+import os
 import pytest
 
 from vllm.multimodal import MULTIMODAL_REGISTRY
 
 from ....conftest import _ImageAssets
 from ...utils import build_model_context
+from ....utils import models_path_prefix
 
 
-@pytest.mark.parametrize("model_id", ["microsoft/Phi-4-multimodal-instruct"])
+@pytest.mark.parametrize("model_id", [os.path.join(models_path_prefix, "microsoft/Phi-4-multimodal-instruct")])
 # yapf: disable
 @pytest.mark.parametrize(
     ("mm_processor_kwargs", "expected_toks_per_img"),

tests/models/multimodal/processing/test_qwen2_vl.py

 # SPDX-License-Identifier: Apache-2.0
 
+import os
 import pytest
 
 from vllm.multimodal import MULTIMODAL_REGISTRY
 
 from ....conftest import _ImageAssets
 from ...utils import build_model_context
+from ....utils import models_path_prefix
 
 
-@pytest.mark.parametrize("model_id", ["Qwen/Qwen2-VL-2B-Instruct"])
+@pytest.mark.parametrize("model_id", [os.path.join(models_path_prefix, "Qwen/Qwen2-VL-2B-Instruct")])
 # yapf: disable
 @pytest.mark.parametrize(
     ("mm_processor_kwargs", "expected_toks_per_img", "expected_pixels_shape"), [

tests/models/multimodal/processing/test_smolvlm.py

 # SPDX-License-Identifier: Apache-2.0
 """Tests for smolvlm's multimodal preprocessing kwargs."""
+import os
 import pytest
 from transformers import SmolVLMConfig
 
@@ -7,9 +8,10 @@ from vllm.multimodal import MULTIMODAL_REGISTRY
 
 from ....conftest import _ImageAssets
 from ...utils import build_model_context
+from ....utils import models_path_prefix
 
 
-@pytest.mark.parametrize("model_id", ["HuggingFaceTB/SmolVLM2-2.2B-Instruct"])
+@pytest.mark.parametrize("model_id", [os.path.join(models_path_prefix, "HuggingFaceTB/SmolVLM2-2.2B-Instruct")])
 # yapf: disable
 @pytest.mark.parametrize(
     ("mm_processor_kwargs", "expected_toks_per_img"),

tests/models/registry.py

@@ -8,7 +8,9 @@ import os
 import pytest
 from packaging.version import Version
 from transformers import __version__ as TRANSFORMERS_VERSION
+# from ..utils import models_path_prefix
 
+models_path_prefix = os.getenv('VLLM_OPTEST_MODELS_PATH') or os.getenv("OPTEST_MODELS_PATH")
 
 
 @dataclass(frozen=True)
@@ -109,8 +111,6 @@ class _HfExamplesInfo:
                 pytest.skip(msg)
 
 
-models_path_prefix = os.getenv('VLLM_OPTEST_MODELS_PATH') or os.getenv("OPTEST_MODELS_PATH")
-
 # yapf: disable
 _TEXT_GENERATION_EXAMPLE_MODELS = {
     # [Decoder-only]

tests/models/test_registry.py

@@ -2,6 +2,7 @@
 
 import warnings
 
+import os
 import pytest
 import torch.cuda
 
@@ -20,6 +21,8 @@ from vllm.platforms import current_platform
 from ..utils import create_new_process_for_each_test
 from .registry import HF_EXAMPLE_MODELS
 
+models_path_prefix = os.getenv('VLLM_OPTEST_MODELS_PATH') or os.getenv("OPTEST_MODELS_PATH")
+
 
 @pytest.mark.parametrize("model_arch", ModelRegistry.get_supported_archs())
 def test_registry_imports(model_arch):
@@ -52,12 +55,12 @@ def test_registry_imports(model_arch):
 
 @create_new_process_for_each_test()
 @pytest.mark.parametrize("model_arch,is_mm,init_cuda,is_ce", [
-    ("LlamaForCausalLM", False, False, False),
-    ("MllamaForConditionalGeneration", True, False, False),
-    ("LlavaForConditionalGeneration", True, True, False),
-    ("BertForSequenceClassification", False, False, True),
-    ("RobertaForSequenceClassification", False, False, True),
-    ("XLMRobertaForSequenceClassification", False, False, True),
+    (os.path.join(models_path_prefix, "LlamaForCausalLM"), False, False, False),
+    (os.path.join(models_path_prefix, "MllamaForConditionalGeneration"), True, False, False),
+    (os.path.join(models_path_prefix, "LlavaForConditionalGeneration"), True, True, False),
+    (os.path.join(models_path_prefix, "BertForSequenceClassification"), False, False, True),
+    (os.path.join(models_path_prefix, "RobertaForSequenceClassification"), False, False, True),
+    (os.path.join(models_path_prefix, "XLMRobertaForSequenceClassification"), False, False, True),
 ])
 def test_registry_model_property(model_arch, is_mm, init_cuda, is_ce):
     assert ModelRegistry.is_multimodal_model(model_arch) is is_mm
@@ -77,9 +80,9 @@ def test_registry_model_property(model_arch, is_mm, init_cuda, is_ce):
 
 @create_new_process_for_each_test()
 @pytest.mark.parametrize("model_arch,is_pp,init_cuda", [
-    ("MLPSpeculatorPreTrainedModel", False, False),
-    ("DeepseekV2ForCausalLM", True, False),
-    ("Qwen2VLForConditionalGeneration", True, True),
+    (os.path.join(models_path_prefix, "MLPSpeculatorPreTrainedModel"), False, False),
+    (os.path.join(models_path_prefix, "DeepseekV2ForCausalLM"), True, False),
+    (os.path.join(models_path_prefix, "Qwen2VLForConditionalGeneration"), True, True),
 ])
 def test_registry_is_pp(model_arch, is_pp, init_cuda):
     assert ModelRegistry.is_pp_supported_model(model_arch) is is_pp