[CI] Skip the pooling models that do not support transformers v4.55 (#22411)

Signed-off-by: wang.yuqi <noooop@126.com>

tests/models/language/pooling/test_embedding.py

@@ -7,7 +7,7 @@ import pytest
 from vllm.config import PoolerConfig
 from vllm.platforms import current_platform
 
-from ...utils import check_embeddings_close
+from ...utils import check_embeddings_close, check_transformers_version
 
 
 @pytest.fixture(autouse=True)
@@ -56,6 +56,9 @@ def test_models(
     model,
     monkeypatch,
 ) -> None:
+    if model == "Alibaba-NLP/gte-Qwen2-1.5B-instruct":
+        check_transformers_version(model, max_transformers_version="4.53.2")
+
     if model == "BAAI/bge-multilingual-gemma2" and current_platform.is_rocm():
         # ROCm Triton FA does not currently support sliding window attention
         # switch to use ROCm CK FA backend

tests/models/language/pooling/test_gte.py

@@ -4,6 +4,7 @@ from typing import Any
 
 import pytest
 
+from ...utils import check_transformers_version
 from .embed_utils import EmbedModelInfo, correctness_test_embed_models
 from .mteb_utils import mteb_test_embed_models
 
@@ -60,6 +61,10 @@ MODELS = [
 @pytest.mark.parametrize("model_info", MODELS)
 def test_embed_models_mteb(hf_runner, vllm_runner,
                            model_info: EmbedModelInfo) -> None:
+    if model_info.name == "Alibaba-NLP/gte-Qwen2-1.5B-instruct":
+        check_transformers_version(model_info.name,
+                                   max_transformers_version="4.53.2")
+
     vllm_extra_kwargs: dict[str, Any] = {}
     if model_info.architecture == "GteNewModel":
         vllm_extra_kwargs["hf_overrides"] = {"architectures": ["GteNewModel"]}
@@ -72,6 +77,10 @@ def test_embed_models_mteb(hf_runner, vllm_runner,
 def test_embed_models_correctness(hf_runner, vllm_runner,
                                   model_info: EmbedModelInfo,
                                   example_prompts) -> None:
+    if model_info.name == "Alibaba-NLP/gte-Qwen2-1.5B-instruct":
+        check_transformers_version(model_info.name,
+                                   max_transformers_version="4.53.2")
+
     vllm_extra_kwargs: dict[str, Any] = {}
     if model_info.architecture == "GteNewModel":
         vllm_extra_kwargs["hf_overrides"] = {"architectures": ["GteNewModel"]}

tests/models/language/pooling/test_reward.py

@@ -10,6 +10,7 @@ from transformers import AutoModel
 from vllm.platforms import current_platform
 
 from ....conftest import HfRunner
+from ...utils import check_transformers_version
 
 
 @pytest.fixture(autouse=True)
@@ -86,6 +87,9 @@ def test_prm_models(
     dtype: str,
     monkeypatch,
 ) -> None:
+    check_transformers_version("Qwen/Qwen2.5-Math-PRM-7B",
+                               max_transformers_version="4.53.2")
+
     if current_platform.is_cpu() and os.environ.get("VLLM_USE_V1", "0") == "0":
         pytest.skip("CPU only supports V1")
 

tests/models/utils.py

@@ -412,3 +412,14 @@ def dummy_hf_overrides(
         })
 
     return hf_config
+
+
+def check_transformers_version(model: str,
+                               min_transformers_version: Optional[str] = None,
+                               max_transformers_version: Optional[str] = None):
+    from .registry import _HfExamplesInfo
+
+    return _HfExamplesInfo(model,
+                           min_transformers_version=min_transformers_version,
+                           max_transformers_version=max_transformers_version
+                           ).check_transformers_version(on_fail="skip")