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 11, 03:45 UTC
[CI] FIx `fixture 'siglip_attention_config' not found` (#31053)
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Lucas Wilkinson committed
Dec 19, 2025 at 22:46 UTC
ff2168bca3a195b835c64a5c9012d7b6a9f34e61
1 file changed
+18
tests/models/multimodal/pooling/conftest.py
new
+18
@@ -0,0 +1,18 @@
1
+# SPDX-License-Identifier: Apache-2.0
2
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
+"""Pytest configuration for vLLM pooling tests."""
4
+
5
+import pytest
6
+
7
+from vllm.platforms import current_platform
8
+
9
+
10
+@pytest.fixture
11
+def siglip_attention_config():
12
+ """Return attention config for SigLIP tests on ROCm.
13
+
14
+ On ROCm, SigLIP tests require FLEX_ATTENTION backend.
15
+ """
16
+ if current_platform.is_rocm():
17
+ return {"backend": "FLEX_ATTENTION"}
18
+ return None