Methods like Linear Attention models, State Space Models like Mamba, Linear RNNs like DeltaNet, and RWKV solve this problem. However ... training and constant-time complexity during inference decoding ...