Review
Open access
Jul 2026
MATRIX AND VECTOR METHODS FOR LANGUAGE-MODEL EFFICIENCY: DISTINGUISHING EMBEDDING COMPRESSION, TOKEN REDUCTION, AND RAG
A structured critical review of matrix and vector methods used to reduce storage and inference costs in language models and retrieval-augmented generation (RAG) systems shows that matrix factorization, adaptive embeddings, and Matryoshka representations compress parameters or vectors but do not automatically decrease t...
Pedro Emílio Amador Salomão, Lívio Almeida Amador
· Nexus Science Review · 0 citations