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 6, 08:15 UTC
[ROCm][CI] Fix granite_speech test for gfx90a by selecting compatible attention backend (#37611)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Mar 20, 2026 at 04:07 UTC
5a4a1795916a7cf3120ab47cc96f663904bca3f0
1 file changed
+5
-1
tests/models/multimodal/generation/test_granite_speech.py
+5
-1
@@ -39,7 +39,11 @@ models = [MODEL_NAME]
39
def granite_speech_attention_config():
40
"""Return attention config for Granite Speech tests on ROCm."""
41
if current_platform.is_rocm():
42
- return {"backend": "ROCM_AITER_FA"}
42
+ from vllm.platforms.rocm import on_mi3xx
43
+
44
+ if on_mi3xx():
45
+ return {"backend": "ROCM_AITER_FA"}
46
+ return {"backend": "TRITON_ATTN"}
47
return None
48
49