Also right now Sol 5.6 Max is super slow but if it were way faster on a chip (like Taalas' Llama 8b demo) then it would be an extreme value multiplier. But the

Also right now Sol 5.6 Max is super slow but if it were way faster on a chip (like Taalas' Llama 8b demo) then it would be an extreme value multiplier. But the model is so large that "baking it onto a chip" doesn't seem straightforward.