I need help with Neural Amp Modeller models.

[rerdavies]

The most recent release of PiPedal delivers a 20% performance improvement when using TooB NAM with large Wavenet models (generally the models that use the most CPU).

The technique I used is broadly applicable, but I have currently limited it to a very specific model type (the model type used by all my most CPU-intensive models). . The Neural Amp Modeller core engine supports a variety of ML algorithms, each with a variety of configuration options. It's a large change to a difficult piece of code, so I am reluctant to allow models with different configurations to run on the new optimized engine until I get a chance to test the code paths that they use. Currently, only large Wavenet models with matching configuration options are passed to the optimised code; the remaining models are handled by the old un-optimised code.

So, what I need you help with: if you have CPU-intensive models whose performance has not improved on the current PiPedal build, could you please let me know what they are, and where you got them if not from Tonehunt.org, so that I can analyse and test the model configurations they are using, and add them to the list of models that will be passed to the new, optimised ML code.

To be clear, I'm not interested in Nano or Feather models with sub-30% CPU use. Performance on those models is fine. I'm interested in the monsters that are using 45%+ cpu use (on a pi 4).