Is Mixed Precision Getting Widely Used?

I would say that there's been a period of exploration the same maybe the last product cycle where we enabled it in our products, but again a little bit sort of to the necessary. There weren't good tools to sort of automate it so it was a way to address point problems. Oh this layer is having a problem quantizing or we need more output resolution hereso we're going to use a higher precision layer which produces higher resolution activations. Now we're introducing automatic mixed precision and bridging that friction points from taking a model that works great on a bench somewhere for data scientists and getting it onto device.