Did you train BART with the input sequence length up to 2048 tokens?

[zhangyao1994]

Hi, in your QMSum paper, the 5.1 Implementation Details section mentioned that you used fairseq library to implement BART model and you truncate the input text to 2048 tokens.
Did you increase the max sequence length to 2048 for the BART model? May I ask how did you implement that?
I saw some suggestions in the following issue and wonder how exactly you implemented your finetuning.
facebookresearch/fairseq#1685