4 ms·RWKV-LM: RNN with transformer-level performance, without using attention5 points by azhenley 4y ago