You can always split up the learning process, for example learn the majority of your tracks on the chorus, then do another learning pass on a track that wasn’t playing, like a saxophone lick in the bridge. The inter-instance mixing will work even when tracks aren’t playing together.
The same idea applies to short, dynamic tracks like hi hats that take a long time to learn, due to not much audio information being present in a track. You can loop one hi hat hit and do a separate learning pass for it.