Basically, think of HTTP load-testers, but for inference. I want to throw generations (parallel) at my inference server and see how it behaves under sustained load and meassure token generation and prefill and figure out the actual storage needs in-use. For HTTP stuff, I was using siege and k6; but
