I've just released a new version of the Python package used for training NAMs!
As usual, the newest version is automatically used in the online trainer; folks making models locally on their own computers can install the update by typing
pip install --upgrade neural-amp-modelerFull release notes are on GitHub.
What's new?
Here are the big ones:
Automatically stop training early
If you just want a model that hits a specific ESR, then there's a new option in the GUI trainer where you can put in e.g. "0.01" and training will stop once the model reaches that accuracy. This is really helpful when training a batch of models some may train more easily than others. Since this is most helpful for batch training, it's only included in the local GUI trainer, not the Colab trainer.
Audition models as you train
One thing I've heard from people queueing up long (e.g. 1000-epoch) training runs is that they wish they could audition the models during training instead of having to wait until the end. To help with that, I've extended the checkpointing to output .nam models:
Simply load one of these into the plugin as you're training to hear how things are sounding.
Download the standardized input audio file from the GUI trainer
It's been tricky to find the standardized audio file for reamping. The Colab trainer has a link, but the GUI trainer hasn't. I've fixed this by adding a button in the upper-right corner that will take you to the file to download.
Manually set the latency compensation in the Colab trainer
The simplified trainers attempt to compensate for the latency in your reamping setup automatically using the "blips" near the start of the standardized reamping file. However, the calculation can sometimes make a mistake. If you know the latency of your reamp setup (or you checked by hand), you can plug in the number (in samples) to Colab here to be sure...
...so that you can be sure it gets it right. As a reminder, this is currently available in the local GUI trainer under "Advanced options":
Enjoy!