LINEAR-TIME SEQUENCE MODELING

Summary from the TLDR AI newsletter

Another state spaces paper with improved scaling and performance. In this case, they borrow ideas from the LSTM and train a 3B parameter model that outperforms larger 7B parameter Transformer models.