vLLM not using all CPU's available on my machine

[bartmch]

I am running vLLM as a docker container, only specifying the model repository and tensor-parallel-size set to 2. The model loads successfully and I can use the /completion endpoint, both GPU's are used my vLLM. The issue is that only 2 CPU's are being used maxing out at 2*100%. I have 24 CPU's on my system. How can I make sure vLLM uses more than 2 CPU's?