[MuseCoco] Request for Checkpoint of the "large" model

[Sonata165]

Dear MuseCoco Researchers,

I really appreciate the effort of your team on this music generation research and the XL checkpoint!

I'm conducting conditional generation research with this model by finetuning from your checkpoint. I've noticed that while the XL model's training speed is acceptable, its inference speed is considerably slow. With two A40 cards, it takes about 6 minutes to infer just a sample with eight bars. However, in my case, I need to do autoregressive decoding in validation to monitor condition awareness. Now it becomes quite challenging given this time cost.

I'm wondering if it's possible for your team to release the checkpoint of the large (203M parameters) model? I believe it would be very helpful to build a prototype with finetuning on the smaller one, where the time overhead should be much more managable.

Thank you in advance! I'm looking foward to your kind reply.

Best regards,
Longshen Ou