Sparsity-Preserving Differentially Private Training of Large Embedding Models | Read Paper on Bytez