nammy

A proof-of-concept trainer for Neural Amp Modeler's WaveNet architectures (A1 and A2), implemented with tinygrad instead of PyTorch.

What's implemented

Both WaveNet architectures, ported faithfully from nam.models.wavenet: the current standard A2 (the default; --arch a2, or a2-lite for the 3-channel variant) and the classic A1 (--arch a1), with NAM's data pipeline and standard learning config around them. Trained models export to .nam and load in the plugin, A2 in a recent version of it.

Not implemented (yet): packed/slimmable training, gated/FiLM variants, MRSTFT loss, pre-emphasis, output loudness normalization, the standardized input-file splits/checks.

Usage

Train from an input/output pair:

$ uv run python -m nammy train input.wav output.wav --epochs 100 --out model.nam

input.wav is the DI/reamp source, output.wav the processed capture; both must share a sample rate and be time-aligned (use --latency <samples> to compensate reamp latency). --out is rewritten every time validation ESR improves, so interrupting a run leaves the best model so far on disk.

Run audio through a trained model (reamp):

$ uv run python -m nammy process model.nam input.wav output.wav

This also loads WaveNet .nam files trained elsewhere.

Backend

By default nammy tries Metal (on macOS), then OpenCL, then the CPU, and uses the first that actually compiles and runs a kernel. Any tinygrad target can be chosen explicitly with --device, accepted by every command:

$ uv run python -m nammy train input.wav output.wav --device CL

With more than one GPU, discrete cards come first: CL is the best one, CL:1 the next, and so on. nammy devices lists them by name and says which work.

GUI

A small Tkinter front end over the same two commands:

$ uv run python -m nammy gui

On Windows you can instead double-click nammy-gui.pyw. The Train tab streams the log, plots validation ESR per epoch, and can stop a run at the next batch boundary; the Process tab reamps a WAV through a .nam. The device picker at the top is shared by both tabs.

Single-file build

For handing the GUI to someone who has a Python but no interest in installing anything:

$ uv run python tools/build_standalone.py     # -> dist/nammy.pyzw

That is a zipapp: one file holding nammy and tinygrad, double-clickable on Windows, needing nothing installed and unpacking nothing.

Windows Defender's Controlled Folder Access may report that it "blocked python.exe from making changes to memory" (event 1127). That is tinygrad's JIT allocating executable memory and it has not stopped a run here; allowing the interpreter under Ransomware protection silences it.

Notes

• On an RX 6800 a 100-epoch run over a 340 s capture pair takes about 13 min. PERFORMANCE.md has the numbers and where the time goes.
• DESIGN.md covers why the backend is probed, why the GUI uses a single worker thread, and how the zipapp build works.
• PACKAGING.md covers the Microsoft Store build, which is an MSIX with an interpreter inside it rather than a zipapp.
• Tests: uv run tests/test_poc.py and uv run tests/test_no_numpy.py.