Today (𝟭𝟭𝗮𝗺-𝟭𝟮:𝟯𝟬𝗽𝗺 𝗛𝗦𝗧 #-𝟰𝟬𝟵) at #ICML2023 we present STEP, a training recipe for structured sparsity from scratch with preconditioned optimization. Stop by and learn about model compression and sparse representation for large models. We love crowded audience!