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 9, 14:45 UTC
[Perf] avoid duplicate mem_get_info() call in get_current_memory_usage (#33064)
Signed-off-by: Paco Xu <paco.xu@daocloud.io>
Paco Xu committed
Jan 27, 2026 at 11:45 UTC
157caf511b5264c1b93610e5dcde04f70fd9d676
1 file changed
+2
-1
vllm/platforms/rocm.py
+2
-1
@@ -561,7 +561,8 @@ class RocmPlatform(Platform):
561
cls, device: torch.types.Device | None = None
562
) -> float:
563
torch.cuda.reset_peak_memory_stats(device)
564
- return torch.cuda.mem_get_info(device)[1] - torch.cuda.mem_get_info(device)[0]
564
+ free_mem, total_mem = torch.cuda.mem_get_info(device)
565
+ return total_mem - free_mem
566
567
@classmethod
568
def get_device_communicator_cls(cls) -> str: