-
Notifications
You must be signed in to change notification settings - Fork 28
Open
Milestone
Description
Is your feature request related to a problem? Please describe
在用国产推理卡进行推理时,发现supervisor 模式下,vllm进程打不满cpu, 导致性能下降。用htop 观察只有core0,64 活跃。
(在xinference-local 模式下,没有这个问题)

Describe the solution you'd like
经过测试,可以打满cpu
### Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.
Additional context
Add any other context or screenshots about the feature request here.
Metadata
Metadata
Assignees
Labels
No labels