The most direct alternative to heavy rankers is the Microsoft MiniLM family. These models distill the knowledge of massive models (like BERT) into a much smaller architecture.
In the landscape of Information Retrieval (IR) and semantic search, the "bigger is better" mentality has dominated for years. However, the industry is shifting toward efficiency. generally refers to a class of lightweight neural ranking models designed to re-rank documents with minimal computational overhead. alternative tinyranker