We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
So it works on GPU but only uses 6-10% of it. I assume you have not implemented batching yet, still a lot of room for improvement here.