SPLADE Model

SPLADE is a family of neural information retrieval models designed to improve the accuracy and efficiency of document retrieval. Current research focuses on enhancing SPLADE's performance, particularly through architectural modifications like separating query and document encoders and employing regularization techniques to reduce latency without significant accuracy loss. These advancements aim to bridge the performance gap between neural and traditional retrieval systems, making advanced retrieval methods more practical for real-world applications requiring both speed and accuracy. The resulting models achieve state-of-the-art zero-shot performance on various benchmarks, demonstrating significant progress in information retrieval.