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, 21:45 UTC
platforms: Fix Ray DP startup crash (#36665)
Signed-off-by: Itay Alroy <ialroy@nvidia.com>
Itay Alroy committed
Mar 11, 2026 at 12:08 UTC
5353c9b0160586cee8413bfcbc1a11ef1076df47
1 file changed
+5
vllm/platforms/interface.py
+5
@@ -638,6 +638,11 @@ class Platform:
638
"""Raises if this request is unsupported on this platform"""
639
640
def __getattr__(self, key: str):
641
+ # Pickle checks dunder methods like __getstate__. If we return None
642
+ # for them, pickle treats it like a real value and tries to call it.
643
+ if key.startswith("__") and key.endswith("__"):
644
+ raise AttributeError(key)
645
+
646
device = getattr(torch, self.device_type, None)
647
if device is not None and hasattr(device, key):
648
attr = getattr(device, key)