QRNN float16 support - see bfarzin’s post.
Anyone has an idea how to change relevant CUDA code to make it work?