So i am working with CRD3 dataset and BART model for fine tuning, i want to pass different turns of input dialogue chunks sequentially like RNN input and then pass that output embedding of encoder to decoder to generate the summary corresponding to those turns, i don’t know how to use seq2seq trainer to do this, can anyone plese help here.