Uses of Class
ai.beestream.jitnn.nn.Module
Packages that use Module
Package
Description
Neural-network modules after PyTorch's
torch.nn: Module and
Sequential, linear and convolution layers, pooling, normalisation,
attention, embeddings, recurrent layers (LSTM, GRU), graph convolution, activations, and data-parallel
training (DataParallel).Complete models: ResNet, MobileNetV2, GPT-2, a Llama-style decoder, BERT, a VAE, DCGAN, U-Net and
character RNNs, and the embedding models word2vec, doc2vec and GloVe.
-
Uses of Module in ai.beestream.jitnn.nn
Subclasses of Module in ai.beestream.jitnn.nnModifier and TypeClassDescriptionfinal classAdaptive average pooling to fixed output sizes (1 to 3 spatial axes), PyTorch's nn.AdaptiveAvgPool1d/2d/3d.final classAdaptive max pooling to fixed output sizes (1 to 3 spatial axes), PyTorch's nn.AdaptiveMaxPool1d/2d/3d.final classResidual connection: relu(main(x) + shortcut(x)) or main(x) + shortcut(x).final classAverage pooling over 1-D windows, PyTorch's nn.AvgPool1d.final classAverage pooling; padding counts in the divisor, as PyTorch's default.final classAverage pooling over 3-D windows, PyTorch's nn.AvgPool3d.final classBatch normalisation over the channels of NCHW (or NC) input, PyTorch defaults: weight 1, bias 0, running mean 0 and variance 1, momentum 0.1, eps 1e-5.final classCELU withalpha(PyTorch's default 1).final class1-D convolution over [N, C, ...] with a cubic kernel, optionally grouped, PyTorch's initialisation: weight and bias uniform in [-1/sqrt(fan_in), 1/sqrt(fan_in)), fan_in = in * kernel^1.final class2-D convolution over NCHW with a square kernel, PyTorch'snn.Conv2dsemantics and initialisation: weight and bias uniform in [-1/sqrt(fan_in), 1/sqrt(fan_in)), fan_in = in/groups * k * k.final class3-D convolution over [N, C, ...] with a cubic kernel, optionally grouped, PyTorch's initialisation: weight and bias uniform in [-1/sqrt(fan_in), 1/sqrt(fan_in)), fan_in = in * kernel^3.final classTransposed 2-D convolution over NCHW with a square kernel, PyTorch'snn.ConvTranspose2dsemantics and initialisation: weight [in, out/groups, k, k] and bias uniform in [-1/sqrt(fan_in), 1/sqrt(fan_in)), fan_in = out/groups * k * k.final classx^3.final classInverted dropout in training mode, identity in evaluation mode.final classELU with slopealphafor negative inputs (PyTorch's default 1).final classA lookup table ofcountvectors of sizedim, applied to INT32 indices.final classEmbeddingBag: bags of INT32 indices (one flat list plus bag offsets) to the sum, mean or max of their embedding rows, PyTorch's nn.EmbeddingBag with standard normal initial weights.final class[N, d1, d2, ...] to [N, d1 * d2 * ...], a view.final class[N, C, H, W] to [N, C], the mean over each feature map (adaptive average pooling to 1x1, flattened).final classGraph convolution (Kipf and Welling, 2017), PyTorch Geometric's GCNConv:  (x W^T) + b, with  the normalised adjacencySparseMatrix.gcn(int, int[], int[])of the graph, given with each call since one layer serves many graphs. x [nodes, in] gives [nodes, out].final classGroupNorm over [N, C, ...] with a per-channel affine (weight 1, bias 0 at the start), PyTorch's nn.GroupNorm.final classA stack of GRU layers over [T, B, I] sequences (time first), optionally bidirectional, PyTorch'snn.GRUequations, gate order, parameter order and initialisation: every weight and bias uniform in [-1/sqrt(H), 1/sqrt(H)).final classHardSigmoid: clamp(x / 6 + 1/2, 0, 1).final classHardSwish: x hardSigmoid(x).final classHardTanh: x clamped to [min, max].final classInstanceNorm over [N, C, ...]: each sample and channel normalised over its positions, PyTorch's nn.InstanceNorm1d/2d/3d without running statistics; affine optional (PyTorch's default: none).final classLayerNorm over the last axis, with weight 1 and bias 0 initially, like PyTorch'snn.LayerNorm.final classLeakyReLU with a fixed negative slope (PyTorch's default 0.01).final classy = x W^T + b with W of shape [out, in].final classLocal response normalisation across channels, PyTorch's nn.LocalResponseNorm.final classA stack of LSTM layers over [T, B, I] sequences (time first), optionally bidirectional, PyTorch'snn.LSTMequations, gate order, parameter order and initialisation: every weight and bias uniform in [-1/sqrt(H), 1/sqrt(H)).final classMax pooling over 1-D windows, PyTorch's nn.MaxPool1d.final classMax pooling over square windows of NCHW input, PyTorch's nn.MaxPool2d.final classMax pooling over 3-D windows, PyTorch's nn.MaxPool3d.final classMish: x tanh(softplus(x)).final classMulti-head self-attention on [B, T, D] inputs, like PyTorch'snn.MultiheadAttentionwith batch_first: one fused q/k/v projection [3D, D],headsheads of D / heads features, and an output projection.final classPadding of the spatial axes: constant, reflect or replicate (PyTorch's nn.ConstantPadNd, ReflectionPadNd, ReplicationPadNd).final classPReLU with a learnable slope, one shared or one per channel, initialised toinit(PyTorch: 0.25).final classReLU: max(x, 0).final classReLU6: min(max(x, 0), 6).final classRMSNorm over the last axis, with weight 1 initially, like PyTorch'snn.RMSNorm.final classRReLU: random slopes in [lower, upper) while training (a fresh seed per call, so a replayed step draws new slopes without recompiling), the mean slope in evaluation.final classSELU with PyTorch's constants.final classApplies its modules in order.final classThe logistic sigmoid 1 / (1 + exp(-x)).final classSoftplus: log(1 + exp(x)).final classSoftsign: x / (1 + |x|).final classBatch normalisation whose training statistics are those of the batch over all data-parallel ranks, as PyTorch's SyncBatchNorm: each rank sums its x and x^2 per channel, one all-reduce gives the global mean and variance, and the backward pass all-reduces their gradients the same way.final classThe hyperbolic tangent.final classKeras's ThresholdedReLU: x where x > theta, else 0.final classNearest or bilinear upsampling of NCHW by a scale factor, PyTorch'snn.Upsample.Methods in ai.beestream.jitnn.nn with type parameters of type ModuleModifier and TypeMethodDescriptionprotected <M extends Module>
MModule.child(M module) Adds a child module; returns it.Methods in ai.beestream.jitnn.nn that return ModuleModifier and TypeMethodDescriptionDataParallel.module()The wrapped replica.Module.train(boolean on) Switches this module and its children to training mode (on) or evaluation mode; returns this module.Constructors in ai.beestream.jitnn.nn with parameters of type ModuleModifierConstructorDescriptionshortcutnull means identity.DataParallel(Module model, int rank, Collective collective) Wraps rankrank's replicamodel; ranks synchronise throughcollective.Sequential(Module... layers) The modules applied in order, each to the previous one's output. -
Uses of Module in ai.beestream.jitnn.nn.models
Subclasses of Module in ai.beestream.jitnn.nn.modelsModifier and TypeClassDescriptionfinal classA BERT-style encoder for masked language modelling: token and learned position embeddings with LayerNorm, post-LayerNorm blocks of bidirectional attention (with an additive padding mask) and a GELU MLP of 4 x dim, and an MLM head (dense, GELU, LayerNorm, then the token embedding as the decoder with its own bias).final classA recurrent language model: token embedding, a stack of LSTM or GRU layers, a linear head over the vocabulary.static final classImages [B, 1, 28, 28] to logits [B, 1].static final classz [B, latent] to images [B, 1, 28, 28] in [-1, 1].final classParagraph vectors (Le and Mikolov, 2014) with negative sampling.final classGloVe (Pennington et al., 2014): word and context vectors with biases fitted to the logarithm of the co-occurrence counts, weighted by f(x) = min(1, (x / xMax)^alpha).final classA GPT-2 style decoder: token and learned position embeddings, pre-LayerNorm blocks of causal multi-head attention and a GELU (tanh form) MLP of 4 x dim, a final LayerNorm, and logits from the token embedding (tied weights).final classA Llama-style decoder: token embedding, pre-RMSNorm blocks of causal attention with rotary position embedding (rotate-half form) and a SwiGLU feed-forward, a final RMSNorm and an untied output projection; no biases.final classA small U-Net (Ronneberger et al., 2015) for 28x28 inputs with two down-sampling levels: double 3x3 convolutions with ReLU, max pooling, a bottleneck, then one level up through a stride-2 transposed convolution and one through bilinear upsampling and a 1x1 convolution, each concatenated with the encoder's features of that size; a 1x1 convolution gives one logit per pixel.final classA fully connected variational autoencoder (Kingma and Welling, 2014), the layout of PyTorch's MNIST example: 784-400 encoder to a Gaussian latent (mean, log-variance), 400-784 decoder with a sigmoid output.final classword2vec (Mikolov et al., 2013) with negative sampling: an input embedding (the word vectors) and an output embedding, initialised like the original C code (input uniform in +-0.5 / dim, output zero).