Is zero-scaling possible in Triton Server?

[kwanwoo02]

I currently proceeded to autoscaling the triton server using minikube, but I want to do zero scaling and together autoscaling.
I want to zero scale my tritonserver. Is it possible?
The zero scaling I mentioned here means that the triton server is in a standby state and does not occupy memory when there is no inference request. Like Kserve

if possible please tell me how

thank you~

[dyastremsky]

I don't believe that's an option. Have you tried unloading all models when you don't need them? That should get you to the smallest memory footprint. You'll then need to load the model again when the next request comes in.