Finetuning on HF using Seq2SeqTrainer.

[varadhbhatnagar]

I need to finetune csebuetnlp/mT5_m2m_crossSum using the Seq2SeqTrainer on HuggingFace. What special symbols need to be inserted during tokenization of input-output pairs?

[abhik1505040]

Hi, We use language-specific bos tokens only for the decoder. The language-to-bos-token mapping is present in the model config. Please have a look at the code snippet given on the model page to see how you can access it. If you want to readily use our scripts to further fine-tune this model on your own data, just replace google/mt5-base with csebuetnlp/mT5_m2m_crossSum on this line