Exporting
Getting the architecture out in other forms.
Press Export, right of the workspace tabs, and choose a format.
| Format | What you get |
|---|---|
| JSON | The design — blocks, connections, configuration — and its analysis, re-importable into NEURAX. |
| TOML | The same design, in a format meant to be hand-edited and diffed. |
| GitHub | Push the architecture straight to a repository. |
The PyTorch project#
Beside those, the Full Project tab hands over the files a training run is made of, exactly as NEURAX writes them into the run — nothing added for the export.
| File | What it is |
|---|---|
model.py | The architecture as real nn.Module classes with a real forward pass: the canvas translated, or code the assistant or you wrote that passed verification. |
train.py | The harness every run uses: reads the data, trains, writes steps and checkpoints beside itself. |
request.json | What the harness is asked to do: input shape, dataset, epochs, batch size, precision. |
The Full Project tab is different: it holds the run's own model.py, train.py and request.json — put beside a dataset, that folder trains exactly as a run started from the studio does.
Page 24 of 33. This guide is in English for now. It is the same guide that is built into the NEURAX desktop app (the ? button, or F1). Found a mistake? Tell us.