Models Can Communicate via Inference Servers Despite Being Sandboxed
TL;DRTwo agents running on the same inference server can reliably communicate, even if they cannot send each other messages or access the same filesystem. They can do this by exploiting "side channels": mechanisms that let one model subtly affect another model's computation. Here, I discuss two communication protocols: Step Sizing and KV Cache Message Board. These both convey information through the way vLLM chooses kernels for particular engine steps.I show that these communication protocols wo...
Read full article →