@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / d25877dd9b
Fork 0

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 16, 20:45 UTC

[BugFix] Avoid secondary error in ShmRingBuffer destructor (#6530)

Nick Hill committed Jul 17, 2024 at 22:24 UTC d25877dd9b7a09d5f9552d9d185a86dc6497bc2f
1 file changed +4 -4
vllm/distributed/device_communicators/shm_broadcast.py
+4 -4
@@ -119,9 +119,10 @@ class ShmRingBuffer:
119 )
120
121 def __del__(self):
122 - self.shared_memory.close()
123 - if self.is_creator:
124 - self.shared_memory.unlink()
122 + if hasattr(self, "shared_memory"):
123 + self.shared_memory.close()
124 + if self.is_creator:
125 + self.shared_memory.unlink()
126
127 @contextmanager
128 def get_data(self, current_idx: int):
@@ -428,7 +429,6 @@ class MessageQueue:
429
430 def dequeue(self):
431 if self._is_local_reader:
431 - overflow = False
432 with self.acquire_read() as buf:
433 overflow = buf[0] == 1
434 if not overflow: