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, 12:15 UTC
[Bugfix][ROCm] Using device_type because on ROCm the API is still torch.cuda (#17601)
Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
Gregory Shtrasberg committed
May 3, 2025 at 01:25 UTC
a92842454ca824ce6fcf356f31e3bf0daf53629b
1 file changed
+2
-2
vllm/platforms/interface.py
+2
-2
@@ -406,12 +406,12 @@ class Platform:
406
"""Raises if this request is unsupported on this platform"""
407
408
def __getattr__(self, key: str):
409
- device = getattr(torch, self.device_name, None)
409
+ device = getattr(torch, self.device_type, None)
410
if device is not None and hasattr(device, key):
411
return getattr(device, key)
412
else:
413
logger.warning("Current platform %s does not have '%s'" \
414
- " attribute.", self.device_name, key)
414
+ " attribute.", self.device_type, key)
415
return None
416
417
@classmethod