A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 04:45 UTC
[CPU][CI] Fallback sliding window to v0 and fix CPU pooling model tests (#19901)
Signed-off-by: jiang1.li <jiang1.li@intel.com>
Li, Jiang committed
Jun 20, 2025 at 23:30 UTC
79f2f1c2a1999d1e7a5202062bad4e115fd9d775
2 files changed
+13
-1
tests/models/language/pooling/test_embedding.py
+6
-1
@@ -1,5 +1,7 @@
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
+import os
4
+
5
import pytest
6
7
from vllm.config import PoolerConfig
@@ -33,7 +35,7 @@ def v1(run_with_both_engines):
35
# To avoid this problem, for now we skip v0 since it will be
36
# deprecated anyway.
37
pytest.param("ssmits/Qwen2-7B-Instruct-embed-base",
36
- marks=[pytest.mark.skip_v0]),
38
+ marks=[pytest.mark.skip_v0, pytest.mark.cpu_model]),
39
# [Encoder-only]
40
pytest.param("BAAI/bge-base-en-v1.5",
41
marks=[
@@ -58,6 +60,9 @@ def test_models(
60
model,
61
monkeypatch,
62
) -> None:
63
+ if model == "intfloat/e5-mistral-7b-instruct" and current_platform.is_cpu(
64
+ ) and os.environ.get("VLLM_USE_V1", "0") == "1":
65
+ pytest.skip("CPU V1 doesn't support sliding window")
66
67
if model == "BAAI/bge-multilingual-gemma2" and current_platform.is_rocm():
68
# ROCm Triton FA does not currently support sliding window attention
vllm/engine/arg_utils.py
+7
@@ -1449,6 +1449,13 @@ class EngineArgs:
1449
model_config=model_config) and _warn_or_fallback(
1450
current_platform.device_name):
1451
return False
1452
+
1453
+ if (current_platform.is_cpu()
1454
+ and model_config.get_sliding_window() is not None):
1455
+ _raise_or_fallback(feature_name="sliding window (CPU backend)",
1456
+ recommend_to_remove=False)
1457
+ return False
1458
+
1459
#############################################################
1460
1461
return True