Package io.github.kirstenali.deepj.layers

Interface Projection

All Superinterfaces:

Layer, Trainable

All Known Implementing Classes:

Linear

public interface Projection
extends Layer

Marker interface for linear projection layers (fully-connected, no activation).

Narrows Layer to layers that map [n × dIn] → [n × dOut] via a learnable weight matrix (and optional bias). Used as the field type for the LM-head slot in DecoderOnlyModel so the constructor rejects anything that is not a projection.

Method Summary

Methods inherited from interface io.github.kirstenali.deepj.layers.Layer

backward, forward

Methods inherited from interface io.github.kirstenali.deepj.training.Trainable

parameters, zeroGrad