I made a local web UI for training NAM amp models — WaveNet/A1 only, for hardware that doesn't support A2
Hey,
I built a small local tool with a web interface for training NAM amp models — the reason is pretty specific:
The problem:
Tone3000 (the most popular online trainer) now only trains A2 (SlimmableContainer) models. But a lot of hardware amp modelers — like the Valeton GP-50 and others — only support the classic WaveNet (A1) architecture. A2 models simply won't load on these devices, even though the file looks valid.
So if you own one of these devices, Tone3000 is no longer an option for you.
Repo: https://github.com/arturksd/NAM-A1-local-trainer
What this tool does:
- Local web UI — no cloud, no account, runs on your machine
- Trains WaveNet (A1) models only, deliberately
- Uses neural-amp-modeler 0.12.2 — newer versions export the 0.7.x
.namformat which many hardware modelers don't support yet. This stays on 0.5.x. - Select input WAV + output WAV, set epochs, pick a save location, click Start
- Live progress bar with time estimate and ESR quality rating
Runs on: macOS, Linux, Windows
Setup takes a few minutes (Python, ffmpeg, nam 0.12.2, Flask) — everything is in the README.
If your hardware modeler doesn't support A2 and you're stuck, this might help. Happy to answer questions!