Problem: We need reliable saving of button weights and to verify how architecture dumping interacts with weight saving. Users also want a UI option to choose whether to save the model architecture alongside weights.
Goals:
- Ensure button weights are persisted correctly when users trigger a save action.
- Verify and, if necessary, fix the architecture dumping logic so that architecture + weights can be saved together or separately.
- Add UI control allowing the user to opt-in/out of saving the architecture along with weights.
Acceptance criteria:
- Saving a button's weights writes a consistent file (format and location documented) and can be loaded back to restore behavior.
- Architecture dumping is verified: when requested, the architecture is exported in a compatible format; when not requested, only weights are saved.
- UI includes a clear toggle "Save architecture with weights" with appropriate defaults and persistence across sessions.
- Backwards compatibility: older weight-only files still load correctly; handle cases where architecture is missing.
- Tests for saving/loading weights-only, architecture+weights, and mismatched pairs.
Implementation notes / suggestions:
- Decide on file formats (e.g., PyTorch .pt/.pth, TensorFlow SavedModel, or JSON + binary weights) and document them.
- If saving architecture, include a metadata manifest describing model version, framework, and save timestamp.
- Expose an API endpoint for programmatic saves (for CLI or automated workflows).
References: user request — "Saving buttons weights and check dumping architecture, allowing also user to set if we wants to save the architecture with weights from UI".
Problem: We need reliable saving of button weights and to verify how architecture dumping interacts with weight saving. Users also want a UI option to choose whether to save the model architecture alongside weights.
Goals:
Acceptance criteria:
Implementation notes / suggestions:
References: user request — "Saving buttons weights and check dumping architecture, allowing also user to set if we wants to save the architecture with weights from UI".