Alexey Buzovkin

I keep coming back to the question of how to make generative models run faster without losing the details. Lately that has meant building TensorRT kernels and testing diffusion decoders on images and video.