I think it is a very interesting architecture w/ the proper CUDA kernel: faster than LSTMs + with customizable recurrent poolings/gates. Are there any plans of adding more recurrent layers, and more specifically QRNNs? Thanks!
I think it is a very interesting architecture w/ the proper CUDA kernel: faster than LSTMs + with customizable recurrent poolings/gates. Are there any plans of adding more recurrent layers, and more specifically QRNNs? Thanks!