How to make inference with Composer model

[meliksahturker]

I am interested in generating multi-track polyphonic music from single latent vector, hence the composer model. My question is how can I use composer model with existing pre-trained models?

I made some experiments with the default model but it seems its input is 25x128 vector, reason of which, I don't know, and I couldn't find a way/arg to use composer model.