Implement a Dropout layer (stochastic mask in training, identity in eval).
Status: scope needs to be worked out — mask storage between forward/backward, RNG choice (XorShift32 vs. rand(), see issue #54), train/eval mode handling, quantization paths, test plan.
Part of the layer set Leo took on after coordination with Jan (alongside Convolution, LayerNorm, MaxPool, AvgPool).
Implement a
Dropoutlayer (stochastic mask in training, identity in eval).Status: scope needs to be worked out — mask storage between forward/backward, RNG choice (XorShift32 vs. rand(), see issue #54), train/eval mode handling, quantization paths, test plan.
Part of the layer set Leo took on after coordination with Jan (alongside Convolution, LayerNorm, MaxPool, AvgPool).