@article{exploitingtransformeractivationsparsity, title = {Exploiting Activation Sparsity with Dense to Dynamic-k Mixture-of-Experts Conversion}, author = {Filip Szatkowski and Bartosz Wójcik and Mikołaj Piórczyński and Simone Scardapane}, year = {2023}, eprint = {2310.04361}, archivePrefix = {arXiv}, url = {https://arxiv.org/abs/2310.04361v4}, }