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 12, 13:30 UTC
[UX] Put CUDA attention backend selection log into one line (#29337)
Signed-off-by: mgoin <mgoin64@gmail.com>
Michael Goin committed
Nov 25, 2025 at 09:46 UTC
dbc3d9991ab0e5adc0db6a8c71c9059268032a14
1 file changed
+2
-4
vllm/platforms/cuda.py
+2
-4
@@ -407,9 +407,6 @@ class CudaPlatformBase(Platform):
407
408
# We have found some valid backends. Select the one with the
409
# highest priority.
410
- logger.info(
411
- "Valid backends: %s", [b[0].name for b in valid_backends_priorities]
412
- )
410
sorted_indices = sorted(
411
range(len(valid_backends_priorities)),
412
key=lambda i: valid_backends_priorities[i][1],
@@ -417,8 +414,9 @@ class CudaPlatformBase(Platform):
414
selected_index = sorted_indices[0]
415
selected_backend = valid_backends_priorities[selected_index][0]
416
logger.info(
420
- "Using %s backend.",
417
+ "Using %s attention backend out of potential backends: %s",
418
selected_backend.name,
419
+ [b[0].name for b in valid_backends_priorities],
420
)
421
422
return selected_backend.get_path()