I have the dataset having each sample size upto 10000 tokens after tokenization, so i want to fine tune the seq2seq pretrained model (CodeT5p) on my own custom dataset . I tried using the chunking method which is given in this ( GitHub - mim-solutions/bert_for_longer_texts: BERT classification model for processing texts longer than 512 tokens. Text is first divided into smaller chunks and after feeding them to BERT, intermediate results are pooled. The implementation allows fine-tuning. ). Can someone suggest me how to fine tune with big sentences and generate the big sentences by fine tuning the pretrained model which takes input embedding size limited 512.