Audio context window

[tomheno]

Hello, thanks for your impressive work on this model and novel reinforcement learning methodology.
I am interested into using it and fine-tune it but I found no notion of the current context window for audio (for conversation) that can be processed.

Could you provide this information ?

Thanks

[runninging]

During training, we use a max duration of 300 seconds for audio input, and training with 64k sequence length