will infinity use all available cpus? or would I rather just give it one cpu and run multiple instances?

[nikcaryo-super]

just trying to deploy an embedding server for sentence transformers models without reinventing the wheel.

[michaelfeil]

Yes - why not try it out?

optimum (onnx) / ct2 / torch are using all cores. Don't be shocked, the default setting in these libs will be all physical cores (aka w/o hyperthreading), which should be a good choice.