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 12, 22:15 UTC
[BugFix] Ray with multiple nodes (#28873)
Signed-off-by: Julien Denize <julien.denize@mistral.ai>
Julien Denize committed
Nov 19, 2025 at 22:20 UTC
cdeec2e6067613c501f82463d54e420097f49750
1 file changed
+8
-8
vllm/v1/worker/gpu_worker.py
+8
-8
@@ -205,14 +205,14 @@ class Worker(WorkerBase):
205
assert self.local_rank < torch.cuda.device_count(), (
206
f"DP adjusted local rank {self.local_rank} is out of bounds. "
207
)
208
- visible_device_count = (
209
- torch.cuda.device_count() if torch.cuda.is_available() else 0
210
- )
211
- assert self.parallel_config.local_world_size <= visible_device_count, (
212
- f"local_world_size ({self.parallel_config.local_world_size}) must be "
213
- f"less than or equal to the number of visible devices "
214
- f"({visible_device_count})."
215
- )
208
+ visible_device_count = (
209
+ torch.cuda.device_count() if torch.cuda.is_available() else 0
210
+ )
211
+ assert self.parallel_config.local_world_size <= visible_device_count, (
212
+ f"local_world_size ({self.parallel_config.local_world_size}) must "
213
+ f"be less than or equal to the number of visible devices "
214
+ f"({visible_device_count})."
215
+ )
216
self.device = torch.device(f"cuda:{self.local_rank}")
217
current_platform.set_device(self.device)
218