Expand description
Auto-generated module
🤖 Generated with SplitRS
Structs§
- Adam
Optimizer - Adam optimizer for a flat parameter vector.
- Atomic
Neural Network - Atomic neural network potential (NNP) with one sub-network per element.
- Attention
Readout - Attention-based graph readout that computes a weighted sum of node features.
- Batch
Norm Layer - Batch normalization layer (inference mode).
- Behler
Parrinello Descriptor - Behler-Parrinello symmetry functions for constructing atomic descriptors.
- Conv1D
Layer - A 1-D convolutional layer operating on a sequence of feature vectors.
- Data
Normalizer - Feature-wise Z-score normalizer.
- Dense
Layer - A single fully-connected (dense) layer with an activation function.
- Dense
Layer64 - A fully-connected layer with f64 weights supporting forward pass and gradient computation for backpropagation.
- Dropout
Layer - Dropout regularisation layer.
- Feed
Forward Net - A sequential feed-forward neural network.
- GnnLayer
- A single graph neural network layer implementing the sum-aggregation message passing update:
- GpuNeural
Buffer - A mock GPU buffer for batched neural network inference.
- Grad
Accumulator - Accumulates gradients from multiple backward passes for mini-batch training.
- Inference
Pipeline - An inference pipeline that chains DenseLayer and BatchNormLayer operations.
- Layer
Norm - Layer normalisation applied to each time step independently.
- Layer
Norm Layer - Layer normalisation (Ba et al., 2016).
- Message
Passing Net - A multi-layer message passing neural network stacking
GnnLayers. - Multi
Head Attention - Multi-head attention module.
- Network
Builder - Convenience builder for standard AANN architectures.
- Neural
Layer - A single fully-connected layer with f64 weights.
- Neural
Network - A sequential feed-forward neural network using f64 precision.
- Positional
Encoding - Sinusoidal positional encoding (Vaswani et al., 2017).
- RnnCell
- A single-step Elman RNN cell:
h_t = activation(W_x * x_t + W_h * h_{t-1} + b). - Transformer
Block - A single transformer encoder block: x → MHA(LayerNorm(x)) + x → FFN(LayerNorm(·)) + ·
- Transformer
Ffn - Position-wise feed-forward network used inside a transformer block.
Enums§
- Activation
Fn - Activation functions for neural network layers.
- Activation
Fn64 - Activation function for f64-precision neural network layers.
- ExtActivation
- Extended activation functions with additional variants for f64 paths.
- Inference
Op - A single operation in the inference pipeline.