[CPU][CI] Fallback sliding window to v0 and fix CPU pooling model tests (#19901)

Signed-off-by: jiang1.li <jiang1.li@intel.com>

tests/models/language/pooling/test_embedding.py

@@ -1,5 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+import os
+
 import pytest
 
 from vllm.config import PoolerConfig
@@ -33,7 +35,7 @@ def v1(run_with_both_engines):
         # To avoid this problem, for now we skip v0 since it will be
         # deprecated anyway.
         pytest.param("ssmits/Qwen2-7B-Instruct-embed-base",
-                     marks=[pytest.mark.skip_v0]),
+                     marks=[pytest.mark.skip_v0, pytest.mark.cpu_model]),
         # [Encoder-only]
         pytest.param("BAAI/bge-base-en-v1.5",
                      marks=[
@@ -58,6 +60,9 @@ def test_models(
     model,
     monkeypatch,
 ) -> None:
+    if model == "intfloat/e5-mistral-7b-instruct" and current_platform.is_cpu(
+    ) and os.environ.get("VLLM_USE_V1", "0") == "1":
+        pytest.skip("CPU V1 doesn't support sliding window")
 
     if model == "BAAI/bge-multilingual-gemma2" and current_platform.is_rocm():
         # ROCm Triton FA does not currently support sliding window attention

vllm/engine/arg_utils.py

@@ -1449,6 +1449,13 @@ class EngineArgs:
                 model_config=model_config) and _warn_or_fallback(
                     current_platform.device_name):
             return False
+
+        if (current_platform.is_cpu()
+                and model_config.get_sliding_window() is not None):
+            _raise_or_fallback(feature_name="sliding window (CPU backend)",
+                               recommend_to_remove=False)
+            return False
+
         #############################################################
 
         return True