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, 07:00 UTC
Add get_device_uuid for rocm (#37694)
Signed-off-by: Tiffany Mintz <Tiffany.Mintz@amd.com>
tmm77 committed
Mar 20, 2026 at 23:33 UTC
1c472f8fe16560f4de7e67554eed12c5a08ee354
1 file changed
+15
vllm/platforms/rocm.py
+15
@@ -28,6 +28,7 @@ try:
28
from amdsmi import (
29
AmdSmiException,
30
amdsmi_get_gpu_asic_info,
31
+ amdsmi_get_gpu_device_uuid,
32
amdsmi_get_processor_handles,
33
amdsmi_init,
34
amdsmi_shut_down,
@@ -608,6 +609,20 @@ class RocmPlatform(Platform):
609
return _ROCM_DEVICE_ID_NAME_MAP[device_name]
610
return asic_info["market_name"]
611
612
+ @classmethod
613
+ @with_amdsmi_context
614
+ def get_device_uuid(cls, device_id: int = 0) -> str:
615
+ try:
616
+ device = amdsmi_get_processor_handles()[device_id]
617
+ except AmdSmiException as error:
618
+ logger.error("amdsmi device query failed ", exc_info=error)
619
+ return ""
620
+ try:
621
+ device_uuid = amdsmi_get_gpu_device_uuid(device)
622
+ except AmdSmiException as error:
623
+ logger.error("amdsmi device uuid query failed ", exc_info=error)
624
+ return device_uuid
625
+
626
@classmethod
627
def get_device_total_memory(cls, device_id: int = 0) -> int:
628
device_props = torch.cuda.get_device_properties(device_id)