TurboSparse: Elite Inference Speed via dReLU Sparsity

Achieve 2-5x faster LLM decoding on RTX 4090 and mobile devices using TurboSparse. Experience 97% parameter sparsity without performance loss.