Default time_transformer_depth

lucidrains / BS-RoFormer

Implementation of Band Split Roformer, SOTA Attention network for music source separation out of ByteDance AI Labs

MIT License

384 stars 13 forks source link

Closed iver56 closed 10 months ago

iver56 commented 10 months ago

In readme you use the following parameters as example:

time_transformer_depth = 1
freq_transformer_depth = 1

However, these are the defaults in the BSRoformer class:

time_transformer_depth = 2
freq_transformer_depth = 2

What would you recommend?

lucidrains commented 10 months ago

this hyperparameter should not matter as much as the main depth