Fastai_v1, adding features

QRNN float16 support - see bfarzin’s post.

Anyone has an idea how to change relevant CUDA code to make it work?