Hi. We recently had MoE models as big as 1T and 1.6T total parameters. My expectation on the proportion between total and active parameters so far was around 10 to 1, that we save on smaller, "actually local" models. However, these new huge models have a much smaller number of active parameters for