Training

[decahedron1]

Support for the Training API.

for (inputs, labels) in dataset {
	let outputs = trainer.step(ort::inputs![inputs]?, ort::inputs![labels]?)?;
	let loss: ArrayView0<f32> = outputs[0].try_extract_tensor::<f32>()?.into_dimensionality().unwrap();
	println!("{}", loss.into_scalar());
	optimizer.step()?;
	optimizer.reset_grad()?;
}

TODO

• Documentation + examples
• Complete API
• HF-like Trainer API
• Add build pipeline 🙃

[codecov]

Codecov Report

Attention: Patch coverage is 5.71429% with 132 lines in your changes missing coverage. Please review.

Project coverage is 44.45%. Comparing base (dc79ade) to head (8a0646c).

Files	Patch %	Lines
src/training/simple.rs	0.00%	84 Missing ⚠️
src/training/trainer.rs	0.00%	48 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #202      +/-   ##
==========================================
- Coverage   47.58%   44.45%   -3.14%     
==========================================
  Files          27       30       +3     
  Lines        1839     1973     +134     
==========================================
+ Hits          875      877       +2     
- Misses        964     1096     +132     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.