3 ms·Faster training, much faster inference, and about half the VRAM usage during inference.by MacsHeadroom 3y agoFaster training, much faster inference, and about half the VRAM usage during inference.