< Back to IRCAM Forum

Different behaviour of nn~ and Rave VST

The output of my model in unconditional prior mode does not match the examples generated during training. Initially, I thought the model might not export correctly in --streaming / --cached modes. However, when I use the nn~ object in Max instead of the VST plugin, the unconditional generation aligns with the examples from the training process.

Based on this observation, I assume the issue may be related to the VST plugin not caching.