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 14, 05:30 UTC
[Minor] Simplify duplicative device check for cuda (#24793)
Signed-off-by: Ziliang Peng <ziliangdotme@gmail.com>
Victor Ziliang Peng committed
Sep 13, 2025 at 11:28 UTC
973c9d01da863cac9c51e8a5c0d390fc84b84fbc
1 file changed
+1
-2
vllm/platforms/cuda.py
+1
-2
@@ -64,8 +64,7 @@ class CudaPlatformBase(Platform):
64
if self.has_device_capability(80):
65
# Ampere and Hopper or later NVIDIA GPUs.
66
return [torch.bfloat16, torch.float16, torch.float32]
67
- elif (not self.has_device_capability(80)
68
- ) and self.has_device_capability(60):
67
+ if self.has_device_capability(60):
68
# Pascal, Volta and Turing NVIDIA GPUs, BF16 is not supported
69
return [torch.float16, torch.float32]
70
# Kepler and Maxwell NVIDIA GPUs, only FP32 is supported,