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 7, 01:30 UTC
[Bugfix] Fix CPU OMP autobind assertion to use local_world_size (#35815)
Signed-off-by: liweiguang <codingpunk@gmail.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Li, Jiang <jiang1.li@intel.com>
Weiguang Li committed
Mar 9, 2026 at 11:07 UTC
43aa3892314f8336f83a9fbe614899ddcf0e1df8
1 file changed
+3
-5
vllm/v1/worker/cpu_worker.py
+3
-5
@@ -144,12 +144,10 @@ class CPUWorker(Worker):
144
allowed_numa_nodes, logical_cpu_list = (
145
CpuPlatform.get_allowed_cpu_core_node_list()
146
)
147
- assert (
148
- len(allowed_numa_nodes) >= self.parallel_config.world_size
149
- or sim_multi_numa_nodes
150
- ), (
147
+ local_world_size = self.parallel_config.local_world_size
148
+ assert len(allowed_numa_nodes) >= local_world_size or sim_multi_numa_nodes, (
149
f"Not enough allowed NUMA nodes to bind threads of "
152
- f"{self.parallel_config.world_size} CPUWorkers. "
150
+ f"{local_world_size} local CPUWorkers. "
151
f"Allowed NUMA nodes are {allowed_numa_nodes}. "
152
"Please try to bind threads manually."
153
)