RT-Lynx: Putting GEMM Sparsity in the Right Place for Diffusion Models
The episode discusses 'RT-Lynx,' a paper detailing how to apply sparsity in diffusion models. The hosts conclude that applying sparsity to activations, rather than static weights, is crucial for maintaining image quality and improving speed. RT-Lynx achieves significant performance gains while preserving fidelity across various architectures through specialized techniques.