Tag
1 article
Learn about SPARSEUP, a 149M-parameter sparse embedding model that sets new benchmarks for efficient retrieval systems. Understand how it leverages logit shift, top-k expansion, and case folding to achieve high performance while maintaining sparsity.