Most people are using something in the llama family for inference. Llama server is my go to. Unsloth guides describe how to configure inference for your model of choice.