5 ms·You Only Cache Once: Decoder-Decoder Architectures for Language Models3 points by reqo 2y agosimonttt 2y agoBasically, it is a hybrid model like Jamba.