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, 11:00 UTC
[V1] Enable TPU V1 backend by default (#17673)
Signed-off-by: mgoin <mgoin64@gmail.com>
Michael Goin committed
May 6, 2025 at 09:49 UTC
d419aa5dc4a519e4c8bdefbf767457dd687d9556
1 file changed
+4
-3
vllm/engine/arg_utils.py
+4
-3
@@ -1354,9 +1354,10 @@ class EngineArgs:
1354
if is_eagle_enabled and _warn_or_fallback("Eagle"):
1355
return False
1356
1357
- # Non-CUDA is supported on V1, but off by default for now.
1358
- not_cuda = not current_platform.is_cuda()
1359
- if not_cuda and _warn_or_fallback( # noqa: SIM103
1357
+ # Non-[CUDA, TPU] may be supported on V1, but off by default for now.
1358
+ v0_hardware = not any(
1359
+ (current_platform.is_cuda(), current_platform.is_tpu()))
1360
+ if v0_hardware and _warn_or_fallback( # noqa: SIM103
1361
current_platform.device_name):
1362
return False
1363
#############################################################