曾明亮

Hi @SkyAndCloud It seems that the length of the token sequence is not a multiple of 8. Did you change some parameters in the scripts?

The error happened in the data loader, so it was probably **not** caused by `--fp16`.

This is the [archive](https://1drv.ms/u/s!Av1IXjAYTqPstx9kydgjLgmSauVt?e=fgnuqi) of the preprocessed dataset `lmd_data_bin`.

Hi @AniketRajpoot - I think prepending class tokens will work. But the genre classes in TOP-MAGD are general and ambiguous, so the condition may be too weak for generation tasks....

Hi @tripathiarpan20 Migrating the code should be easy. But migrating the pre-trained checkpoints is tricky. The MusicBERT plugin overrides a few methods of the original RoBERTa-related classes. If the original...