Quick 'n dirty tool for distilling NAM models.
Make sure you're in a Python environment that has neural-amp-modeler installed.
- Run
./init.sh - Then
./build.sh - Then
./distill.sh <model.nam>
That easy!
- Run
docker build -t nam-distillery . - Then
docker run -it --rm nam-distillery
Note: If you have an NVIDIA GPU, add the --gpus all flag to the above docker run command.
This will drop you into the shell of your container. You can use curl <web.address> to grab a model, or scp or something to grab it from your local filesystem.
Once you have your desired model in your container, you can run ./distill.sh <model.nam>
Currently configured to use Edward Payne's reamping CLI to distill arbitrary NAM models to the "pico" model definition designed by GuitarML.
You can provide your own model definition in nam_full_configs/model.json.
You can modify training parameters in nam_full_configs/learn.json
By default, NeuralAmpModelerReamping uses fast_tanh, for better accuracy you can comment this line out in the source code before building.