Add a small dense multi-layer-perceptron forward pass (linear layers plus activations) for evaluating exported policy weights on embedded targets. Inference only, no training. Planned after 1.0.
Done when: a sample network matches a reference forward pass.
Depends on #62
Add a small dense multi-layer-perceptron forward pass (linear layers plus activations) for evaluating exported policy weights on embedded targets. Inference only, no training. Planned after 1.0.
Done when: a sample network matches a reference forward pass.
Depends on #62