LayerHelper<T>

Helpers & Utilities · AiDotNet.Helpers

Provides helper methods for creating various neural network layer configurations.

How It Works

This class contains factory methods that create pre-configured sets of neural network layers for common architectures like standard feed-forward networks, CNNs, ResNets, and more.

Methods

Method Summary
DenseLayer<T> CreateDefaultABINetBranchHead(int charsetSize) Creates a character-probability head for one of ABINet’s branches.
IEnumerable<ILayer<T>> CreateAudioVAEDecoderLayers(int melChannels, int latentChannels, int baseChannels, int[] channelMultipliers) Creates decoder layers for the AudioVAE.
IEnumerable<ILayer<T>> CreateAudioVAEEncoderLayers(int melChannels, int latentChannels, int baseChannels, int[] channelMultipliers) Creates encoder layers for the AudioVAE.
IEnumerable<ILayer<T>> CreateAudioVisualCorrespondenceLayers(int embeddingDimension, int numEncoderLayers, int numAttentionHeads) Creates layers for the Audio-Visual Correspondence network (L3-Net).
IEnumerable<ILayer<T>> CreateAudioVisualEventLocalizationLayers(int inputSize, int embeddingDimension, int numEncoderLayers, int numCategories, int audioEmbeddingFullyConnectedWidth, int audioEmbeddingSize) Creates layers for the Audio-Visual Event Localization network.
IEnumerable<ILayer<T>> CreateBasicVSRPlusPlusLayers(int channels, int height, int width, int numFeatures, int scaleFactor, int numResidualBlocks, int numPropagations, int numLevels, int deformGroups, int growthChannels, double residualScale) Creates layers for the BasicVSRPlusPlus video super-resolution model.
IEnumerable<ILayer<T>> CreateBiomedParseDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the BiomedParse model.
IEnumerable<ILayer<T>> CreateBiomedParseEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the BiomedParse model.
IEnumerable<ILayer<T>> CreateBlip2Layers(int imageSize, int channels, int patchSize, int visionHiddenDim, int qformerHiddenDim, int numQformerLayers, int numHeads, int vocabularySize, int embeddingDimension, int lmHiddenDim, int numLmDecoderLayers, int maxSequenceLength) Creates layers for the BLIP-2 native mode.
IEnumerable<ILayer<T>> CreateBlipLayers(int imageSize, int patchSize, int hiddenDim, int embeddingDimension, int numLayers, int numDecoderLayers, int numHeads, int mlpDim, int vocabularySize) Creates layers for the BLIP native mode.
IEnumerable<ILayer<T>> CreateCATSegDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the CATSeg model.
IEnumerable<ILayer<T>> CreateCATSegEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the CATSeg model.
IEnumerable<ILayer<T>> CreateCUPSDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the CUPS model.
IEnumerable<ILayer<T>> CreateCUPSEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the CUPS model.
IEnumerable<ILayer<T>> CreateConcertoDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the Concerto model.
IEnumerable<ILayer<T>> CreateConcertoEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the Concerto model.
IEnumerable<ILayer<T>> CreateControlNetEncoderLayers(int conditionChannels, int baseChannels, int[] channelMultipliers, int imageSize) Creates layers for the ControlNet encoder.
IEnumerable<ILayer<T>> CreateDCCRNLayers(int numStages, int numLstmLayers, int lstmHiddenDim, int projectionDim, bool useComplexMask) Creates the DCCRN layers that live in the model’s layer list: the per-stage BatchNorms, the bottleneck LSTM stack, the LSTM-output projection, and the mask activation.
IEnumerable<ILayer<T>> CreateDEVADecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the DEVA model.
IEnumerable<ILayer<T>> CreateDEVAEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate, bool useGroupNormalization) Creates encoder layers for the DEVA model.
IEnumerable<ILayer<T>> CreateDeepFilterNetLayers(int numErbBands, int hiddenDim, int numGruLayers, int dfBins, int dfOrder) Creates DeepFilterNet encoder, GRU, deep filtering, and decoder layers.
IEnumerable<ILayer<T>> CreateDefaultABINetFusionLayers(int visionDim, int charsetSize) Creates ABINet’s fusion branch: the ITERATIVE refinement stack and the final character head.
IEnumerable<ILayer<T>> CreateDefaultABINetLanguageLayers(int charsetSize, int visionDim, int languageDim, int languageLayers) Creates ABINet’s language model (LM): the gradient barrier followed by the Bidirectional Cloze Network.
IEnumerable<ILayer<T>> CreateDefaultABINetLayers(int imageWidth, int imageHeight, int visionDim, int languageDim, int numIterations, int charsetSize) Creates default ABINet (Autonomous, Bidirectional, Iterative) layers as one flat chain.
IEnumerable<ILayer<T>> CreateDefaultABINetVisionLayers(int imageWidth, int imageHeight, int visionDim, int numLayers) Creates default ABINet (Autonomous, Bidirectional, Iterative) layers.
IEnumerable<ILayer<T>> CreateDefaultACEStepLayers(int latentDim, int uNetDim, int numUNetLayers, int textEncoderDim, double dropoutRate, int inputChannels) Creates default layers for ACE-Step accelerated music generation.
IEnumerable<ILayer<T>> CreateDefaultALIGNLayers(int visionEmbeddingDim, int textEmbeddingDim, int projectionDim, int numVisionLayers, int numTextLayers, int numTextHeads, double dropoutRate) Creates default layers for the ALIGN model (EfficientNet CNN vision encoder + text transformer).
IEnumerable<ILayer<T>> CreateDefaultAlphaFactorLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int numFactors, int numAssets, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultAlphaFactorLayers(int numFeatures, int hiddenDimension, int numFactors, int numAssets, double dropoutRate) Creates default layers for an AlphaFactorModel.
IEnumerable<ILayer<T>> CreateDefaultAnimateDiffLayers(int inputChannels, int inputHeight, int inputWidth, int numLayers, int numFrames) Creates layers for an AnimateDiff motion module that adds temporal coherence.
IEnumerable<ILayer<T>> CreateDefaultAttentionLayers(NeuralNetworkArchitecture<T> architecture) Creates a default set of attention-based layers for transformer-style architectures.
IEnumerable<ILayer<T>> CreateDefaultAudioDiffusionLayers(int melChannels, int latentChannels, int baseChannels) Creates default audio diffusion layers for mel spectrogram-based audio generation.
IEnumerable<ILayer<T>> CreateDefaultAudioFlamingo2Layers(int audioEncoderDim, int llmHiddenDim, int numPerceiverLayers, int numPerceiverTokens, double dropoutRate) Creates default layers for Audio Flamingo 2 multimodal audio-language model.
IEnumerable<ILayer<T>> CreateDefaultAudioGenLayers(int textHiddenDim, int lmHiddenDim, int numLmLayers, int numHeads, int numCodebooks, int codebookSize, int maxTextLength, int maxAudioTokens, double dropoutRate) Creates default AudioGen layers for text-to-audio generation.
IEnumerable<ILayer<T>> CreateDefaultAudioLDMClassifierLayers(int numMels, int latentDim, int classifierDim, int numClassifierLayers, int numClasses, double dropoutRate) Creates default layers for AudioLDM Classifier using diffusion features.
IEnumerable<ILayer<T>> CreateDefaultAudioLDMLayers(int textHiddenDim, int latentDim, int unetChannels, int numResBlocks, int[] attentionResolutions, int numHeads, int numMels, int maxTextLength, double dropoutRate) Creates default AudioLDM layers for text-to-audio generation using latent diffusion.
IEnumerable<ILayer<T>> CreateDefaultAudioLMLayers(int semanticDim, int numSemanticLayers, int numSemanticHeads, int semanticVocabSize, double dropoutRate) Creates default layers for AudioLM (semantic stage).
IEnumerable<ILayer<T>> CreateDefaultAudioMAELayers(int patchFeatureSize, int embeddingDim, int numEncoderLayers, int numAttentionHeads, int feedForwardDim, int numClasses, int maxSequenceLength, double dropoutRate) Creates default Audio-MAE (Masked Autoencoders for Audio) layers for classification.
IEnumerable<ILayer<T>> CreateDefaultAudioSepLayers(int clapEmbeddingDim, int separationDim, int numSeparationLayers, int numHeads, int[] encoderChannels, int numClasses, int numMels, double dropoutRate) Creates default layers for the AudioSep model.
IEnumerable<ILayer<T>> CreateDefaultAudioSuperResolutionLayers(int numBlocks, int channels, int kernelSize) Creates the 1-D convolutional U-Net blocks for the Audio Super-Resolution model of Kuleshov, Enam & Ermon (2017), “Audio Super-Resolution using Neural Networks”.
IEnumerable<ILayer<T>> CreateDefaultAutoDiffTabDenoiserLayers(int inputDim, int outputDim, int[] hiddenDims, int timestepEmbeddingDim, double dropoutRate) Creates default denoiser layers for AutoDiffTab automated diffusion.
IEnumerable<ILayer<T>> CreateDefaultAutoEncoderLayers(NeuralNetworkArchitecture<T> architecture, int latentSize) Creates a default autoencoder neural network architecture.
IEnumerable<ILayer<T>> CreateDefaultAutoIntLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int embeddingDimension, int numHeads, int numLayers, int numClasses, double dropoutRate, IVectorActivationFunction<T> hiddenVectorActivation) Creates default layers for an AutoInt (Automatic Feature Interaction) model.
IEnumerable<ILayer<T>> CreateDefaultBASICLayers(int visionEmbeddingDim, int textEmbeddingDim, int projectionDim, int numVisionLayers, int numTextLayers, int numVisionHeads, int numTextHeads, double dropoutRate) Creates default layers for the BASIC model (CoAtNet hybrid CNN-Transformer vision + text transformer).
IEnumerable<ILayer<T>> CreateDefaultBEATsLayers(int patchFeatureSize, int embeddingDim, int numEncoderLayers, int numAttentionHeads, int feedForwardDim, int numClasses, int maxSequenceLength, double dropoutRate) Creates the default layer configuration for a BEATs (Audio Pre-Training with Acoustic Tokenizers) audio event detection and classification model.
IEnumerable<ILayer<T>> CreateDefaultBGELayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultBGELayers(int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Creates default layers for a BGE (BAAI General Embedding) model.
IEnumerable<ILayer<T>> CreateDefaultBSRoFormerLayers(int numBands, int bandEmbeddingDim, int transformerDim, int numTransformerLayers, int numAttentionHeads, int feedForwardDim, int numStems, int numFreqBins, double dropoutRate) Creates default BS-RoFormer layers for band-split music source separation.
IEnumerable<ILayer<T>> CreateDefaultBSVDLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures, int temporalFrames, int numUNetStages, int shiftedChannelRatio) Creates layers for a video denoising model (U-Net style temporal denoiser).
IEnumerable<ILayer<T>> CreateDefaultBandSplitRNNEnhancerLayers(int numBands, int bandRnnHiddenSize, int numRnnLayers, int fusionDim, int numFreqBins, double dropoutRate) Creates default layers for Band-Split RNN enhancer (Luo and Yu, 2023).
IEnumerable<ILayer<T>> CreateDefaultBandSplitRNNSeparationLayers(int numBands, int bandRnnHiddenSize, int numBandRnnLayers, int sequenceRnnHiddenSize, int numSequenceRnnLayers, int fusionDim, int numStems, int numFreqBins, double dropoutRate) Creates default layers for the BandSplitRNN source separation model.
IEnumerable<ILayer<T>> CreateDefaultBasicPitchLayers(int numHarmonicBins, int encoderFilters, int numEncoderLayers, int numMidiNotes, double dropoutRate) Creates default Basic Pitch multi-pitch detection layers.
IEnumerable<ILayer<T>> CreateDefaultBayesianNeuralNetworkLayers(NeuralNetworkArchitecture<T> architecture) Creates a default configuration of layers for a Bayesian neural network (Bayes-by-Backprop style).
IEnumerable<ILayer<T>> CreateDefaultBiLSTMCRFLayers(int embeddingDimension, int hiddenDimension, int numLabels, int numLSTMLayers, int maxSequenceLength, double dropoutRate, bool useCharEmbeddings, int charEmbeddingDimension, int charHiddenDimension, bool useCRF) Creates the default layer stack for a BiLSTM-CRF sequence labeling NER model.
IEnumerable<ILayer<T>> CreateDefaultBiaffineNERLayers(int hiddenDimension, int numAttentionHeads, int numTransformerLayers, int intermediateDimension, int spanEmbeddingDimension, int numLabels, double dropoutRate, int ffnnDepth, int biLstmHiddenSize, int biLstmLayers, double biLstmDropout, double embeddingsDropout) Creates the default layer stack for a span-based NER model (SpERT, BiaffineNER, PURE).
IEnumerable<ILayer<T>> CreateDefaultBlackLittermanNeuralLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int numAssets, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultBlackLittermanNeuralLayers(int numFeatures, int hiddenDimension, int numAssets, double dropoutRate) Creates default layers for a BlackLittermanNeural model.
IEnumerable<ILayer<T>> CreateDefaultBlip2Layers(int imageSize, int channels, int patchSize, int vocabularySize, int embeddingDimension, int qformerHiddenDim, int visionHiddenDim, int lmHiddenDim, int numQformerLayers, int numHeads, int numLmDecoderLayers, int maxSequenceLength) Creates default layers for a BLIP-2 neural network.
IEnumerable<ILayer<T>> CreateDefaultBloombergGPTLayers(NeuralNetworkArchitecture<T> architecture, int maxSequenceLength, int vocabularySize, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultBloombergGPTLayers(NeuralNetworkArchitecture<T> architecture, int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numClasses, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultBloombergGPTLayers(int maxSequenceLength, int vocabularySize, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Creates the default layer configuration for BloombergGPT.
IEnumerable<ILayer<T>> CreateDefaultBloombergGPTLayers(int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numClasses, double dropoutRate) Creates the default layer configuration for BloombergGPT-style model.
IEnumerable<ILayer<T>> CreateDefaultBranchformerLayers(int encoderDim, int numLayers, int numAttentionHeads, int cgmlpDim, int numMels, int vocabSize, double dropoutRate, int maxSequenceLength) Creates default layers for a Branchformer encoder with parallel attention + cgMLP branches.
IEnumerable<ILayer<T>> CreateDefaultBridgeFusionLayers(int visionDim, int textDim, int fusionDim, int numVisionLayers, int numTextLayers, int numBridgeLayers, int numHeads, double dropoutRate, int patchSize) Creates default layers for BridgeTower with bridge connections between vision and text encoder layers.
IEnumerable<ILayer<T>> CreateDefaultCCDMLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int hiddenDimension, int numLayers, int numHeads, double dropout, int embeddingLayers, double mlpRatio, double attentionDropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultCCDMLayers(int contextLength, int forecastHorizon, int hiddenDimension, int numLayers, int numHeads, double dropout, int embeddingLayers, double mlpRatio, double attentionDropout) Creates the denoiser of CCDM, the Channel-aware Contrastive Conditional Diffusion model (Li, Chen and Xiong 2024, arXiv:2410.02168), in the fixed order CCDM binds it.
IEnumerable<ILayer<T>> CreateDefaultCIFEncoderLayers(int encoderDim, int numLayers, int numAttentionHeads, int feedForwardDim, int numMels, int vocabSize, double cifThreshold, double dropoutRate) Creates default layers for a CTC decoder ASR model.
IEnumerable<ILayer<T>> CreateDefaultCLAPAudioEncoderLayers(int numMelBands, int specSize, int patchSize, int embedDim, int[] stageDepths, int[] stageHeads, int windowSize, int mlpRatio, double dropPathRate, bool featureFusion, int projectionDim, int specAugmentTimeDropWidth, int specAugmentTimeStripes, int specAugmentFreqDropWidth, int specAugmentFreqStripes) Creates LAION-CLAP’s audio tower (Wu et al.
IEnumerable<ILayer<T>> CreateDefaultCLAPTextEncoderLayers(int vocabSize, int maxPositions, int textHiddenDim, int textEncoderLayers, int textEncoderHeads, int intermediateDim, int paddingTokenId, double dropoutRate, double layerNormEpsilon, int projectionDim) Creates LAION-CLAP’s text tower: RoBERTa (Liu et al.
IEnumerable<ILayer<T>> CreateDefaultCLIPTextLayers(int vocabSize, int maxSeqLen, int hiddenSize, int numLayers, int numHeads) Paper-faithful CLIP text encoder stack (Radford et al., ICML 2021).
IEnumerable<ILayer<T>> CreateDefaultCMGANLayers(int numFreqBins, int conformerDim, int numConformerLayers, int numAttentionHeads, double dropoutRate) Creates default CMGAN layers for conformer-based speech enhancement.
IEnumerable<ILayer<T>> CreateDefaultCNNBiLSTMCRFLayers(int embeddingDimension, int hiddenDimension, int numLabels, int numLSTMLayers, int maxSequenceLength, double dropoutRate, int charEmbeddingDimension, int charCNNFilters, int charCNNKernelSize) Creates the default layer stack for a CNN-BiLSTM-CRF Named Entity Recognition model.
IEnumerable<ILayer<T>> CreateDefaultCNNLayers(NeuralNetworkArchitecture<T> architecture, int convLayerCount, int filterCount, int kernelSize, int denseLayerCount, int denseLayerSize, int outputSize) Creates a Convolutional Neural Network (CNN) with configurable layers.
IEnumerable<ILayer<T>> CreateDefaultCREPELayers(int frameSize, int capacityMultiplier, int numBins, double dropoutRate) Creates default CREPE pitch detection layers.
IEnumerable<ILayer<T>> CreateDefaultCRNNEventDetectorLayers(int[] cnnChannels, int rnnHiddenSize, int numRNNLayers, int numClasses, int numMels, double dropoutRate) Creates default layers for the CRNN Sound Event Detector.
IEnumerable<ILayer<T>> CreateDefaultCSDILayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int numFeatures, int hiddenDim, int numResidualLayers, int numHeads, int timeEmbeddingDim, int featureEmbeddingDim) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultCSDILayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int numFeatures, int hiddenDimension, int numResidualLayers, int numHeads, int timeEmbeddingDim, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultCSDILayers(int sequenceLength, int numFeatures, int hiddenDim, int numResidualLayers, int numHeads, int timeEmbeddingDim, int featureEmbeddingDim) Creates default layers for the CSDI score-based diffusion model.
IEnumerable<ILayer<T>> CreateDefaultCSDILayers(int sequenceLength, int numFeatures, int hiddenDimension, int numResidualLayers, int numHeads, int timeEmbeddingDim, double dropout) Creates default layers for the CSDI score-based diffusion model.
IEnumerable<ILayer<T>> CreateDefaultCTABGANPlusDiscriminatorLayers(int inputDim, int[] hiddenDims, double dropoutRate) Creates default discriminator layers for a CTAB-GAN+ discriminator.
IEnumerable<ILayer<T>> CreateDefaultCTABGANPlusGeneratorLayers(int inputDim, int outputDim, int[] hiddenDims) Creates default generator layers for a CTAB-GAN+ generator.
IEnumerable<ILayer<T>> CreateDefaultCTCDecoderLayers(int encoderDim, int numLayers, int numAttentionHeads, int feedForwardDim, int numMels, int vocabSize, double dropoutRate, int maxSequenceLength)
IEnumerable<ILayer<T>> CreateDefaultCTGANDiscriminatorLayers(int inputDim, int[] hiddenDims, double dropoutRate) Creates default discriminator layers for a CTGAN discriminator.
IEnumerable<ILayer<T>> CreateDefaultCTGANGeneratorLayers(int inputDim, int outputDim, int[] hiddenDims) Creates default generator layers for a CTGAN generator with residual connections.
IEnumerable<ILayer<T>> CreateDefaultCanaryLayers(int encoderDim, int numEncoderLayers, int decoderDim, int numDecoderLayers, int numHeads, int vocabSize, double dropoutRate) Creates default layers for Canary multilingual ASR/ST model.
IEnumerable<ILayer<T>> CreateDefaultCapsuleNetworkLayers(NeuralNetworkArchitecture<T> architecture) Creates a default capsule network architecture.
IEnumerable<ILayer<T>> CreateDefaultChatGLM3TextLayers(int vocabSize, int maxSeqLen, int hiddenSize, int numLayers, int numHeads, int numKvHeads, double ropeTheta) Paper-faithful ChatGLM3 text encoder stack (Zeng et al.
IEnumerable<ILayer<T>> CreateDefaultChronosBoltLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int encoderHiddenDim, int decoderHiddenDim, int numEncoderLayers, int numDecoderLayers, int numHeads, int numQuantiles, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultChronosBoltLayers(int contextLength, int forecastHorizon, int patchLength, int encoderHiddenDim, int decoderHiddenDim, int numEncoderLayers, int numDecoderLayers, int numHeads, int numQuantiles, double dropout) Creates the default layers for Chronos-Bolt (encoder-decoder with direct quantile output).
IEnumerable<ILayer<T>> CreateDefaultCitrinetLayers(int encoderDim, int numMegaBlocks, int subBlocksPerBlock, int numMels, int vocabSize, double dropoutRate, int seReductionRatio) Builds a paper-faithful NVIDIA Citrinet encoder (Majumdar et al., 2021, https://arxiv.org/abs/2104.01721): a 1-D time-channel separable convolutional CTC model with squeeze-and-excitation and residual mega-blocks, operating on channels-fi…
IEnumerable<ILayer<T>> CreateDefaultClipLayers(NeuralNetworkArchitecture<T> architecture, int projectionDim) Creates default layers for CLIP-style multimodal networks.
IEnumerable<ILayer<T>> CreateDefaultCogVideoLayers(int inputChannels, int inputHeight, int inputWidth, int embedDim, int numLayers, int numFrames) Creates layers for a CogVideo text-to-video generation model.
IEnumerable<ILayer<T>> CreateDefaultColBERTLayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Creates default layers for a ColBERT model whose encoder is hiddenDimension wide.
IEnumerable<ILayer<T>> CreateDefaultColBERTLayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim, int hiddenDimension) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultColBERTLayers(int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Creates default layers for a ColBERT (Contextualized Late Interaction over BERT) model with a BERT-base (768-wide) encoder.
IEnumerable<ILayer<T>> CreateDefaultColBERTLayers(int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim, int hiddenDimension) Creates default layers for a ColBERT model whose encoder is hiddenDimension wide.
IEnumerable<ILayer<T>> CreateDefaultConformerFPLayers(int numMels, int hiddenDim, int numLayers, int numAttentionHeads, int feedForwardDim, int embeddingDim, double dropoutRate) Creates layers for ConformerFP (Conformer-based fingerprinting).
IEnumerable<ILayer<T>> CreateDefaultConformerLayers(int encoderDim, int numLayers, int numAttentionHeads, int feedForwardExpansionFactor, int numMels, int vocabSize, double dropoutRate, int maxSequenceLength, int convKernelSize) Creates default layers for the Conformer ASR model (Gulati et al., 2020).
IEnumerable<ILayer<T>> CreateDefaultConformerTransducerLayers(int encoderDim, int numEncoderLayers, int numAttentionHeads, int feedForwardExpansionFactor, int predictionDim, int jointDim, int numMels, int vocabSize, double dropoutRate, int maxSequenceLength) Creates default layers for a Conformer with RNN-T (Transducer) decoder.
IEnumerable<ILayer<T>> CreateDefaultContextNetLayers(int numBlocks, int numSubBlocks, int numMels, int vocabSize, int kernelSize, int squeezeExcitationRatio, double widthScaling, double dropoutRate) Creates default layers for a deep 1D-CNN based CTC model (Jasper/QuartzNet style).
IEnumerable<ILayer<T>> CreateDefaultCopulaGANDiscriminatorLayers(int inputDim, int[] hiddenDims, double dropoutRate) Creates default discriminator layers for a CopulaGAN discriminator.
IEnumerable<ILayer<T>> CreateDefaultCopulaGANGeneratorLayers(int inputDim, int outputDim, int[] hiddenDims) Creates default generator layers for a CopulaGAN generator.
IEnumerable<ILayer<T>> CreateDefaultCrossAttentionResamplerVLMLayers(int visionDim, int resamplerDim, int decoderDim, int numVisionLayers, int numResamplerLayers, int numDecoderLayers, int numHeads, double dropoutRate, int patchSize) Creates layers for Qwen-VL-style VLMs using cross-attention resampler with learned queries.
IEnumerable<ILayer<T>> CreateDefaultCrossModalFusionLayers(int visionDim, int textDim, int fusionDim, int numRelationshipLayers, int numTextLayers, int numCrossModalLayers, int numHeads, double dropoutRate) Creates default layers for cross-modal fusion models (LXMERT).
IEnumerable<ILayer<T>> CreateDefaultCrossformerLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int predictionHorizon, int numFeatures, int segmentLength, int modelDimension, int numHeads, int numLayers, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultCrossformerLayers(int sequenceLength, int predictionHorizon, int numFeatures, int segmentLength, int modelDimension, int numHeads, int numLayers, double dropout) Creates default layers for the Crossformer (Cross-Dimension Transformer) architecture.
IEnumerable<ILayer<T>> CreateDefaultCutieLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures) Creates layers for a Cutie video object segmentation model.
IEnumerable<ILayer<T>> CreateDefaultDCRNNLayers(NeuralNetworkArchitecture<T> architecture, int numNodes, int numFeatures, int hiddenDimension, int numEncoderLayers, int numDecoderLayers, int forecastHorizon, int diffusionSteps, double[] forwardTransition, double[] backwardTransition, string filterType) Creates the default layers for a DCRNN (Diffusion Convolutional Recurrent Neural Network).
IEnumerable<ILayer<T>> CreateDefaultDIFRINTLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures, int numIterations) Creates default layers for DIFRINT video stabilization.
IEnumerable<ILayer<T>> CreateDefaultDNCLayers(NeuralNetworkArchitecture<T> architecture, int controllerSize, int memoryWordSize, int readHeads, int interfaceSize) Creates a default Differentiable Neural Computer (DNC) with pre-configured layers.
IEnumerable<ILayer<T>> CreateDefaultDPCTGANDiscriminatorLayers(int inputDim, int[] hiddenDims, double dropoutRate) Creates default discriminator layers for a DP-CTGAN discriminator.
IEnumerable<ILayer<T>> CreateDefaultDPCTGANGeneratorLayers(int inputDim, int outputDim, int[] hiddenDims) Creates default generator layers for a DP-CTGAN generator.
IEnumerable<ILayer<T>> CreateDefaultDannaSepLayers(int encoderDim, int numDualPathBlocks, int chunkSize, int numHeads, int numSources, int numFreqBins, double dropoutRate) Creates default layers for Danna-Sep dual-path music source separation.
IEnumerable<ILayer<T>> CreateDefaultData2Vec2Layers(int hiddenDim, int numLayers, int numHeads, int feedForwardDim, double dropoutRate) Creates default layers for data2vec 2.0.
IEnumerable<ILayer<T>> CreateDefaultDecoderOnlyVisionLayers(int patchDim, int decoderDim, int numDecoderLayers, int numHeads, double dropoutRate) Creates layers for decoder-only vision models with no separate vision encoder.
IEnumerable<ILayer<T>> CreateDefaultDeepBeliefNetworkLayers(NeuralNetworkArchitecture<T> architecture, IReadOnlyList<int> hiddenLayerSizes) Creates a default Deep Belief Network (DBN) with pre-configured layers.
IEnumerable<ILayer<T>> CreateDefaultDeepBoltzmannMachineLayers(NeuralNetworkArchitecture<T> architecture, IReadOnlyList<int> hiddenLayerSizes) Creates default layers for a Deep Boltzmann Machine (DBM).
IEnumerable<ILayer<T>> CreateDefaultDeepCNNCTCLayers(int encoderDim, int numBlocks, int numSubBlocks, int numMels, int vocabSize, double dropoutRate)
IEnumerable<ILayer<T>> CreateDefaultDeepFactorLayers(NeuralNetworkArchitecture<T> architecture, int lookbackWindow, int forecastHorizon, int numFeatures, int numFactors, int factorHiddenDim, int localHiddenDim, int numFactorLayers, int numLocalLayers, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultDeepFactorLayers(int lookbackWindow, int forecastHorizon, int numFeatures, int numFactors, int factorHiddenDim, int localHiddenDim, int numFactorLayers, int numLocalLayers, double dropout) Creates the default layer configuration for a DeepFactor (Deep Factor Model).
IEnumerable<ILayer<T>> CreateDefaultDeepPortfolioLayers(NeuralNetworkArchitecture<T> architecture, int numAssets) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultDeepPortfolioLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int numAssets, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultDeepPortfolioLayers(int numAssets) Creates the default layer configuration for a Deep Portfolio Management model.
IEnumerable<ILayer<T>> CreateDefaultDeepPortfolioLayers(int numFeatures, int hiddenDimension, int numAssets, double dropoutRate) Creates default layers for a DeepPortfolioManager model.
IEnumerable<ILayer<T>> CreateDefaultDeepQNetworkLayers(NeuralNetworkArchitecture<T> architecture) Creates a default Deep Q-Network (DQN) with pre-configured layers for reinforcement learning.
IEnumerable<ILayer<T>> CreateDefaultDeepRitzLayers(NeuralNetworkArchitecture<T> architecture, int hiddenLayerCount, int hiddenLayerSize) Creates default layers for the Deep Ritz Method network.
IEnumerable<ILayer<T>> CreateDefaultDeepStateLayers(NeuralNetworkArchitecture<T> architecture, int lookbackWindow, int forecastHorizon, int numFeatures, int stateDimension, int hiddenDimension, int numRnnLayers, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultDeepStateLayers(int lookbackWindow, int forecastHorizon, int numFeatures, int stateDimension, int hiddenDimension, int numRnnLayers, double dropout) Creates the default layer configuration for a DeepState (Deep State Space) model.
IEnumerable<ILayer<T>> CreateDefaultDemucsNoiseLayers(int hiddenChannels, int depth, int lstmHiddenSize, int numLSTMLayers, int channelGrowth, double dropoutRate) Creates default layers for the Demucs Noise model.
IEnumerable<ILayer<T>> CreateDefaultDenseNetLayers(DenseNetConfiguration configuration) Creates default layers for a DenseNet network based on the specified configuration.
IEnumerable<ILayer<T>> CreateDefaultDenseNetLayers(NeuralNetworkArchitecture<T> architecture, DenseNetConfiguration configuration) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultDepthAnythingV2Layers(int inputChannels, int inputHeight, int inputWidth, int numFeatures, int numEncoderBlocks, int patchSize) Creates default layers for Depth Anything V2 monocular depth estimation model.
IEnumerable<ILayer<T>> CreateDefaultDiTLayers(int hiddenDim, int numLayers, int numHeads, int patchSize, int imageSize, int numClasses) Creates default DiT (Document Image Transformer) layers.
IEnumerable<ILayer<T>> CreateDefaultDiTLayers(int latentChannels, int hiddenSize, int numLayers, int numHeads, int patchSize) Creates default DiT (Diffusion Transformer) layers with AdaLN-Zero conditioning.
IEnumerable<ILayer<T>> CreateDefaultDiffusionTSLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int forecastHorizon, int numFeatures, int hiddenDim, int trendHiddenDim, int seasonalHiddenDim) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultDiffusionTSLayers(int sequenceLength, int forecastHorizon, int numFeatures, int hiddenDim, int trendHiddenDim, int seasonalHiddenDim) Creates the default layer configuration for DiffusionTS (Interpretable Diffusion for Time Series).
IEnumerable<ILayer<T>> CreateDefaultDistilledT5TextLayers(int vocabSize, int hiddenSize, int numLayers, int numHeads, int numRelativePositionBuckets, int relativePositionMaxDistance) Paper-faithful DistilledT5 text encoder stack.
IEnumerable<ILayer<T>> CreateDefaultDocBankLayers(int imageSize, int backboneChannels, int numClasses, int hiddenDim) Creates default layers for DocBank page segmentation model.
IEnumerable<ILayer<T>> CreateDefaultDocGCNLayers(int inputDim, int hiddenDim, int numGCNLayers, int numClasses) Creates default DocGCN (Document Graph Convolutional Network) layers.
IEnumerable<ILayer<T>> CreateDefaultDocOwlLayers(int visionDim, int textDim, int visionLayers, int textLayers, int numHeads, int vocabSize, int visionNumHeads, int maxSequenceLength) Creates default DocOwl (mPLUG-DocOwl) layers for document understanding.
IEnumerable<ILayer<T>> CreateDefaultDocumentOCRLayers(int visionDim, int decoderDim, int numVisionLayers, int numDecoderLayers, int numHeads, int maxDocLength, double dropoutRate) Creates layers for document understanding VLMs with reading-order attention.
IEnumerable<ILayer<T>> CreateDefaultDualStreamFusionLayers(int visionDim, int textDim, int fusionDim, int numVisionLayers, int numTextLayers, int numFusionLayers, int numHeads, double dropoutRate, int? numVisionHeads, int? numTextHeads, int? numFusionHeads, int? visionIntermediateDim, int? textIntermediateDim, int? fusionIntermediateDim) Creates default layers for dual-stream vision-language fusion models (ViLBERT, METER).
IEnumerable<ILayer<T>> CreateDefaultEATLayers(int patchFeatureSize, int embeddingDim, int numEncoderLayers, int numAttentionHeads, int feedForwardDim, int numClasses, int maxSequenceLength, double dropoutRate) Creates default EAT (Efficient Audio Transformer) layers following the paper architecture.
IEnumerable<ILayer<T>> CreateDefaultECAPATDNNLanguageIdentifierLayers(NeuralNetworkArchitecture<T> architecture, int numMels, int tdnnChannels, int embeddingDimension, int numLanguages, int[] dilations, int[] kernelSizes, int res2NetScale, int seChannels, int attentionChannels) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultECAPATDNNLanguageIdentifierLayers(int numMels, int tdnnChannels, int embeddingDimension, int numLanguages, int[] dilations, int[] kernelSizes, int res2NetScale, int seChannels, int attentionChannels) Creates default ECAPA-TDNN layers for spoken language identification.
IEnumerable<ILayer<T>> CreateDefaultECAPATDNNSpeakerLayers(int numMels, int channels, int embeddingDim, int numBlocks, int poolingDim, int seBottleneckDim, double dropoutRate, int res2NetScale, int attentionChannels) Creates default ECAPA-TDNN speaker embedding layers.
IEnumerable<ILayer<T>> CreateDefaultEDVRLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures, int numFrames, int numGroups, int numBlocks) Creates default layers for EDVR video restoration.
IEnumerable<ILayer<T>> CreateDefaultELMLayers(NeuralNetworkArchitecture<T> architecture, int hiddenLayerSize) Creates default layers for an Extreme Learning Machine (ELM) neural network.
IEnumerable<ILayer<T>> CreateDefaultESNLayers(int inputSize, int outputSize, int reservoirSize, double spectralRadius, double sparsity) Creates a default Echo State Network (ESN) with pre-configured layers.
IEnumerable<ILayer<T>> CreateDefaultETSformerLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int predictionHorizon, int numFeatures, int modelDimension, int feedForwardDim, int numEncoderLayers, int numDecoderLayers, int numHeads, double dropout) Creates default layers for the ETSformer (Exponential Smoothing Transformer) architecture.
IEnumerable<ILayer<T>> CreateDefaultEditingInstructionLayers(int visionDim, int instructionDim, int editingDim, int numVisionLayers, int numEditingLayers, int numHeads, double dropoutRate) Creates layers for instruction-conditioned image editing VLMs.
IEnumerable<ILayer<T>> CreateDefaultEfficientConformerLayers(int encoderDim, int numLayers, int numAttentionHeads, int feedForwardExpansionFactor, int convKernelSize, int downsamplingFactor, int attentionGroupSize, int numMels, int vocabSize, double dropoutRate, int maxSequenceLength, bool useLayerNormalization) Creates the native EfficientConformer CTC encoder from Burchi and Vielzeuf (2021): three Conformer stages with 120/168/240-width scaling, progressive temporal downsampling, grouped attention in the first stage, and a linear CTC head.
IEnumerable<ILayer<T>> CreateDefaultEmotion2VecLayers(int numMels, int transformerDim, int numTransformerLayers, int numAttentionHeads, int feedForwardDim, int numClasses, double dropoutRate) Creates default emotion2vec layers for speech emotion recognition.
IEnumerable<ILayer<T>> CreateDefaultEncoderDecoderVLMLayers(int visionDim, int decoderDim, int numVisionLayers, int numDecoderLayers, int numHeads, double dropoutRate) Creates default layers for encoder-decoder generative VLMs (GIT, CoCa, PaLI, PaLI-X, PaLI-3).
IEnumerable<ILayer<T>> CreateDefaultFDYSEDLayers(int[] cnnChannels, int embeddingDim, int numFrequencyGroups, int rnnHiddenSize, int numRNNLayers, int numClasses, int numMels, double dropoutRate) Creates default layers for the FDY-SED (Frequency Dynamic SED) model.
IEnumerable<ILayer<T>> CreateDefaultFEDformerLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int predictionHorizon, int numFeatures, int numEncoderLayers, int numDecoderLayers, int numHeads, int modelDimension, int feedForwardDimension) Creates the default layer configuration for a FEDformer (Frequency Enhanced Decomposed Transformer) model.
IEnumerable<ILayer<T>> CreateDefaultFILMLayers(NeuralNetworkArchitecture<T> architecture, int height, int width, int channels, int numScales, int numFeatures) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultFILMLayers(int height, int width, int channels, int numScales, int numFeatures) Creates default layers for a FILM (Frame Interpolation for Large Motion) model.
IEnumerable<ILayer<T>> CreateDefaultFLAVRLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures, int numInputFrames) Creates default layers for FLAVR frame interpolation.
IEnumerable<ILayer<T>> CreateDefaultFRCRNLayers(int encoderChannels, int numStages, int lstmHiddenSize, int numFreqBins, double dropoutRate) Creates default layers for FRCRN (Zhao et al., 2022).
IEnumerable<ILayer<T>> CreateDefaultFTTransformerLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int embeddingDimension, int numHeads, int numLayers, int numClasses, double dropoutRate) Creates default layers for an FT-Transformer (Feature Tokenizer + Transformer) model.
IEnumerable<ILayer<T>> CreateDefaultFactorVAELayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int latentDimension, int numFactors, double dropoutRate, int numAssets) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultFactorVAELayers(int numFeatures, int hiddenDimension, int latentDimension, int numFactors, double dropoutRate, int numAssets) Creates default layers for a FactorVAE model.
IEnumerable<ILayer<T>> CreateDefaultFastConformerLayers(int encoderDim, int numLayers, int numHeads, int feedForwardDim, int convKernelSize, int downsampleFactor, int numMels, int vocabSize, double dropoutRate) Creates default layers for Fast Conformer (8x downsampled Conformer).
IEnumerable<ILayer<T>> CreateDefaultFastDVDNetLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures, int numInputFrames) Creates default layers for FastDVDNet video denoising.
IEnumerable<ILayer<T>> CreateDefaultFastTextLayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int bucketSize, int embeddingDimension) Creates default layers for a FastText model.
IEnumerable<ILayer<T>> CreateDefaultFeedForwardLayers(NeuralNetworkArchitecture<T> architecture, IReadOnlyList<int> hiddenLayerSizes) Creates default layers for a feed-forward neural network with the given hidden-layer widths.
IEnumerable<ILayer<T>> CreateDefaultFeedForwardLayers(NeuralNetworkArchitecture<T> architecture, int hiddenLayerCount, int hiddenLayerSize) Creates default layers for a feed-forward neural network.
IEnumerable<ILayer<T>> CreateDefaultFinBERTLayers(NeuralNetworkArchitecture<T> architecture, int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numSentimentClasses, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultFinBERTLayers(int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numSentimentClasses, double dropoutRate) Creates default layers for FinBERT (Financial BERT) model.
IEnumerable<ILayer<T>> CreateDefaultFinBERTToneLayers(NeuralNetworkArchitecture<T> architecture, int maxSequenceLength, int vocabularySize, int numClasses, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Creates the default layer configuration for FinBERT-Tone (sentiment-focused FinBERT).
IEnumerable<ILayer<T>> CreateDefaultFinBERTToneLayers(NeuralNetworkArchitecture<T> architecture, int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numToneClasses, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultFinBERTToneLayers(int maxSequenceLength, int vocabularySize, int numClasses, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Creates the default layer configuration for FinBERT-tone.
IEnumerable<ILayer<T>> CreateDefaultFinBERTToneLayers(int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numToneClasses, double dropoutRate) Creates the default layer configuration for FinBERT-Tone (sentiment-focused FinBERT).
IEnumerable<ILayer<T>> CreateDefaultFinDiffDenoiserLayers(int inputDim, int outputDim, int[] hiddenDims) Creates default denoiser MLP layers for a FinDiff financial diffusion generator.
IEnumerable<ILayer<T>> CreateDefaultFinDiffTimestepProjectionLayers(int timestepEmbDim) Creates default timestep projection layers for a FinDiff generator.
IEnumerable<ILayer<T>> CreateDefaultFinGPTLayers(NeuralNetworkArchitecture<T> architecture, int maxSequenceLength, int vocabularySize, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultFinGPTLayers(NeuralNetworkArchitecture<T> architecture, int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numClasses, double dropoutRate) Creates the default layer configuration for FinGPT (Financial GPT).
IEnumerable<ILayer<T>> CreateDefaultFinGPTLayers(int maxSequenceLength, int vocabularySize, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Creates the default layer configuration for FinGPT.
IEnumerable<ILayer<T>> CreateDefaultFinMALayers(NeuralNetworkArchitecture<T> architecture, int maxSequenceLength, int vocabularySize, int numAgents, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Creates the default layer configuration for FinMA (Financial Multi-Agent).
IEnumerable<ILayer<T>> CreateDefaultFinMALayers(NeuralNetworkArchitecture<T> architecture, int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numClasses, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultFinMALayers(int maxSequenceLength, int vocabularySize, int numAgents, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Creates the default layer configuration for FinMA.
IEnumerable<ILayer<T>> CreateDefaultFinMALayers(int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numClasses, double dropoutRate) Creates the default layer configuration for FinMA (Financial Multi-Agent).
IEnumerable<ILayer<T>> CreateDefaultFinancialBERTLayers(NeuralNetworkArchitecture<T> architecture, int maxSequenceLength, int vocabularySize, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Creates the default layer configuration for FinancialBERT.
IEnumerable<ILayer<T>> CreateDefaultFinancialBERTLayers(NeuralNetworkArchitecture<T> architecture, int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numClasses, double dropoutRate, string taskType) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultFinancialBERTLayers(int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numClasses, double dropoutRate, string taskType) Creates the default layer configuration for FinancialBERT (domain-adapted financial BERT).
IEnumerable<ILayer<T>> CreateDefaultFlowFormerLayers(int inputChannels, int inputHeight, int inputWidth, int embedDim, int numLayers) Creates default layers for FlowFormer optical flow estimation.
IEnumerable<ILayer<T>> CreateDefaultFlowStateLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int stateDim, int hiddenDim, int numLayers, double dropout, int ssmRank, bool useDiscretization) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultFlowStateLayers(int contextLength, int forecastHorizon, int stateDim, int hiddenDim, int numLayers, double dropout, int ssmRank, bool useDiscretization) Creates the default layers for FlowState (SSM-based foundation model).
IEnumerable<ILayer<T>> CreateDefaultFoundationASRLayers(int featureDim, int encoderDim, int numLayers, int numAttentionHeads, int feedForwardDim, int vocabSize, double dropoutRate, int maxSequenceLength) Creates default layers for a foundation model (wav2vec2/HuBERT/WavLM) fine-tuned for ASR.
IEnumerable<ILayer<T>> CreateDefaultFoundationModelLayers(int featureEncoderDim, int hiddenDim, int numLayers, int numAttentionHeads, int feedForwardDim, double dropoutRate, int maxSequenceLength) Creates default layers for self-supervised speech foundation models (HuBERT, WavLM, wav2vec 2.0).
IEnumerable<ILayer<T>> CreateDefaultFourierNeuralOperatorLayers(NeuralNetworkArchitecture<T> architecture, int[] spatialDimensions, int numFourierLayers, int hiddenChannels, int numModes) Creates default layers for a Fourier Neural Operator (FNO).
IEnumerable<ILayer<T>> CreateDefaultFrameInterpolationLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures) Creates layers for a frame interpolation model (FILM/RIFE-style).
IEnumerable<ILayer<T>> CreateDefaultFullSubNetPlusLayers(int numFreqBins, int fullBandHiddenSize, int subBandHiddenSize, int fullBandLayers, int subBandLayers, double dropoutRate) Creates default FullSubNet+ layers for speech enhancement.
IEnumerable<ILayer<T>> CreateDefaultGANDALFLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int outputSize, int numTrees, int treeDepth, int gatingHiddenDim, int numGatingLayers, int leafDimension, double temperature, double initScale, bool useBatchNorm, double dropoutRate) Creates default layers for a GANDALF (Gated Additive Neural Decision Forest) model.
IEnumerable<ILayer<T>> CreateDefaultGNNLayers(NeuralNetworkArchitecture<T> architecture, int hiddenSize, int numLayers, double dropoutRate, bool useBias) Creates default layers for a Graph Neural Network (GNN).
IEnumerable<ILayer<T>> CreateDefaultGOGGLEDecoderLayers(int latentDim, int outputDim, int[] hiddenDims) Creates default decoder layers for GOGGLE graph-based generation.
IEnumerable<ILayer<T>> CreateDefaultGOGGLEEncoderLayers(int numFeatures, int latentDim, int hiddenDim, int numGCNLayers) Creates default GNN encoder layers for GOGGLE graph-based generation.
IEnumerable<ILayer<T>> CreateDefaultGPT4TSLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numLayers, int numHeads, double dropout) Creates default layers for the GPT4TS model (frozen GPT-2 backbone with task heads).
IEnumerable<ILayer<T>> CreateDefaultGRULayers(NeuralNetworkArchitecture<T> architecture) Creates a default Gated Recurrent Unit (GRU) neural network layer configuration.
IEnumerable<ILayer<T>> CreateDefaultGemmaTextLayers(int vocabSize, int maxSeqLen, int hiddenSize, int numLayers, int numHeads, double ropeTheta) Paper-faithful Gemma text encoder stack (Gemma Team 2024).
IEnumerable<ILayer<T>> CreateDefaultGenreClassifierLayers(int numMels, int hiddenDim, int numClasses, int maxFrames, int numAttentionLayers, double dropoutRate) Creates default genre classification layers.
IEnumerable<ILayer<T>> CreateDefaultGloVeLayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int embeddingDimension) Creates default layers for a GloVe (Global Vectors) model.
IEnumerable<ILayer<T>> CreateDefaultGraFPrintLayers(int numMels, int gnnHiddenDim, int numGnnLayers, int numAttentionHeads, int embeddingDim, double dropoutRate, int kNeighbors, int peakFilters, int peakStride, int peakKernelSize, int encoderEmbeddingDim, int projectionExpansion) Creates the native GraFPrint encoder and projection head from the official implementation.
IEnumerable<ILayer<T>> CreateDefaultGraphAttentionLayers(NeuralNetworkArchitecture<T> architecture, int numHeads, int numLayers, double dropoutRate, int hiddenSize) Creates default layers for a Graph Attention Network (GAT).
IEnumerable<ILayer<T>> CreateDefaultGraphAttentionPortfolioLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int numHeads, int numAssets, int sequenceLength, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultGraphAttentionPortfolioLayers(int numFeatures, int hiddenDimension, int numHeads, int numAssets, int sequenceLength, double dropoutRate) Creates default layers for a GraphAttentionPortfolio (arXiv:2407.15532).
IEnumerable<ILayer<T>> CreateDefaultGraphClassificationLayers(NeuralNetworkArchitecture<T> architecture, int hiddenDim, int embeddingDim, int numGnnLayers, double dropoutRate) Creates default layers for a Graph Classification model.
IEnumerable<ILayer<T>> CreateDefaultGraphGenerationLayers(NeuralNetworkArchitecture<T> architecture, int hiddenDim, int numEncoderLayers) Creates default layers for a Graph Generation model (VGAE encoder).
IEnumerable<ILayer<T>> CreateDefaultGraphIsomorphismLayers(NeuralNetworkArchitecture<T> architecture, int mlpHiddenDim, int numLayers, bool learnEpsilon, double initialEpsilon) Creates default layers for a Graph Isomorphism Network (GIN).
IEnumerable<ILayer<T>> CreateDefaultGraphSAGELayers(NeuralNetworkArchitecture<T> architecture, SAGEAggregatorType aggregatorType, int numLayers, bool normalize) Creates default layers for a GraphSAGE (Graph Sample and Aggregate) Network.
IEnumerable<ILayer<T>> CreateDefaultGraphWaveNetLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int forecastHorizon, int numNodes, int numFeatures, int residualChannels, int skipChannels, int endChannels, int numBlocks, int layersPerBlock) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultGraphWaveNetLayers(int sequenceLength, int forecastHorizon, int numNodes, int numFeatures, int residualChannels, int skipChannels, int endChannels, int numBlocks, int layersPerBlock) Creates default layers for GraphWaveNet (Graph WaveNet for spatial-temporal modeling).
IEnumerable<ILayer<T>> CreateDefaultGroundingDetectionLayers(int visionDim, int textDim, int fusionDim, int detectionDim, int numVisionLayers, int numFusionLayers, int numDetectionLayers, int numHeads, double dropoutRate) Creates layers for grounding/detection VLMs with cross-modal feature fusion.
IEnumerable<ILayer<T>> CreateDefaultHTDemucsLayers(int numFreqBins, int transformerDim, int numTransformerLayers, int numAttentionHeads, int numStems, double dropoutRate) Creates default HTDemucs layers for hybrid transformer music source separation.
IEnumerable<ILayer<T>> CreateDefaultHTMLayers(NeuralNetworkArchitecture<T> architecture, int columnCount, int cellsPerColumn, double sparsityThreshold) Creates a default Hierarchical Temporal Memory (HTM) neural network layer configuration.
IEnumerable<ILayer<T>> CreateDefaultHTSATLayers(int patchFeatureSize, int embeddingDim, int numStages, int[] numLayersPerStage, int numClasses, int maxSequenceLength, double dropoutRate) Creates default HTS-AT (Hierarchical Token-Semantic Audio Transformer) layers.
IEnumerable<ILayer<T>> CreateDefaultHamiltonianLayers(NeuralNetworkArchitecture<T> architecture, int hiddenLayerCount, int hiddenLayerSize) Creates default layers for a Hamiltonian Neural Network.
IEnumerable<ILayer<T>> CreateDefaultHierarchicalRiskParityLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int numAssets, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultHierarchicalRiskParityLayers(int numFeatures, int hiddenDimension, int numAssets, double dropoutRate) Creates default layers for a HierarchicalRiskParity model.
IEnumerable<ILayer<T>> CreateDefaultHippoLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int modelDim, int stateDim, int numLayers, double dropoutRate, bool useNormalization, int numFeatures, int memorySize, string hippoMethod, string discretizationMethod, int initialTime, double timeStep, double timescaleMin, double timescaleMax, bool useGate) Creates default layers for the HiPPO (High-order Polynomial Projection Operators) architecture.
IEnumerable<ILayer<T>> CreateDefaultHuBERTSERLayers(int numMels, int transformerDim, int numTransformerLayers, int numAttentionHeads, int feedForwardDim, int classifierHiddenDim, int numClasses, double dropoutRate) Creates default HuBERT-SER layers for speech emotion recognition.
IEnumerable<ILayer<T>> CreateDefaultITransformerLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int predictionHorizon, int numFeatures, int numLayers, int numHeads, int modelDimension, int feedForwardDimension) Creates the default layer configuration for an iTransformer (Inverted Transformer) model.
IEnumerable<ILayer<T>> CreateDefaultInfographicVQALayers(int imageSize, int visionDim, int textDim, int fusionDim, int visionLayers, int fusionLayers, int numHeads, int vocabSize) Creates default InfographicVQA layers for infographic understanding.
IEnumerable<ILayer<T>> CreateDefaultInstructorLayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultInstructorLayers(int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Creates default layers for an Instructor/E5 (Instruction-Tuned) embedding model.
IEnumerable<ILayer<T>> CreateDefaultInternVideo2Layers(int inputChannels, int inputHeight, int inputWidth, int embedDim, int numEncoderLayers, int patchSize) Creates layers for an InternVideo2-style video understanding model.
IEnumerable<ILayer<T>> CreateDefaultInvestLMLayers(NeuralNetworkArchitecture<T> architecture, int maxSequenceLength, int vocabularySize, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultInvestLMLayers(NeuralNetworkArchitecture<T> architecture, int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numClasses, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultInvestLMLayers(int maxSequenceLength, int vocabularySize, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Creates the default layer configuration for InvestLM.
IEnumerable<ILayer<T>> CreateDefaultInvestLMLayers(int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numClasses, double dropoutRate) Creates the default layer configuration for InvestLM (Investment Language Model).
IEnumerable<ILayer<T>> CreateDefaultKairosLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int[] patchSizes, int hiddenDim, int numLayers, int numHeads, int intermediateSize, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultKairosLayers(int contextLength, int forecastHorizon, int[] patchSizes, int hiddenDim, int numLayers, int numHeads, int intermediateSize, double dropout) Creates the default layers for Kairos (Mixture-of-Size Encoder).
IEnumerable<ILayer<T>> CreateDefaultKeywordSpottingLayers(NeuralNetworkArchitecture<T> architecture, int hiddenLayerCount, int hiddenLayerSize, int outputLabelCount) Creates the feed-forward Deep KWS network from Chen, Parada, and Heigold (ICASSP 2014).
IEnumerable<ILayer<T>> CreateDefaultKronosLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numLayers, int numHeads, int intermediateSize, int numCandlestickFeatures, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultKronosLayers(int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numLayers, int numHeads, int intermediateSize, int numCandlestickFeatures, double dropout) Creates default layers for the Kronos financial market foundation model.
IEnumerable<ILayer<T>> CreateDefaultLLMTimeLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int hiddenDimension, int numLayers, int numHeads, double dropout) Creates default layers for the LLM-Time model (LLM-based zero-shot forecasting).
IEnumerable<ILayer<T>> CreateDefaultLLaVAMLPProjectorLayers(int visionDim, int mlpIntermediateDim, int decoderDim, int numVisionLayers, int numDecoderLayers, int numHeads, double dropoutRate, int patchSize, int inputChannels) Creates layers for LLaVA-style VLMs using a 2-layer MLP cross-modal connector.
IEnumerable<ILayer<T>> CreateDefaultLSMLayers(NeuralNetworkArchitecture<T> architecture, int reservoirSize, double connectionProbability, double spectralRadius, double inputScaling, double leakingRate) Creates a default configuration of layers for a Liquid State Machine (LSM) neural network.
IEnumerable<ILayer<T>> CreateDefaultLSTMCRFLayers(int embeddingDimension, int hiddenDimension, int numLabels, int numLSTMLayers, int maxSequenceLength, double dropoutRate) Creates the default layer stack for an LSTM-CRF Named Entity Recognition model.
IEnumerable<ILayer<T>> CreateDefaultLSTMNetworkLayers(NeuralNetworkArchitecture<T> architecture) Creates a default configuration of layers for a Long Short-Term Memory (LSTM) neural network.
IEnumerable<ILayer<T>> CreateDefaultLSTNetLayers(NeuralNetworkArchitecture<T> architecture, int lookbackWindow, int forecastHorizon, int numFeatures, int convolutionFilters, int convolutionKernelSize, int hiddenRecurrentSize, int hiddenSkipSize, int skipPeriod, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultLSTNetLayers(int lookbackWindow, int forecastHorizon, int numFeatures, int convolutionFilters, int convolutionKernelSize, int hiddenRecurrentSize, int hiddenSkipSize, int skipPeriod, double dropout) Creates the default layers for an LSTNet (Long Short-Term Time-series Network) model.
IEnumerable<ILayer<T>> CreateDefaultLagLlamaLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int numFeatures, int numLags, int hiddenDim, int numLayers, int numHeads, int intermediateSize, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultLagLlamaLayers(int contextLength, int forecastHorizon, int numFeatures, int numLags, int hiddenDim, int numLayers, int numHeads, int intermediateSize, double dropout) Creates default layer configuration for Lag-Llama time series foundation model.
IEnumerable<ILayer<T>> CreateDefaultLagrangianLayers(NeuralNetworkArchitecture<T> architecture, int hiddenLayerCount, int hiddenLayerSize) Creates default layers for a Lagrangian Neural Network.
IEnumerable<ILayer<T>> CreateDefaultLayers(NeuralNetworkArchitecture<T> architecture, IReadOnlyList<int> hiddenLayerSizes, int outputSize) Creates a feed-forward network whose hidden layers have the given widths, in order.
IEnumerable<ILayer<T>> CreateDefaultLayers(NeuralNetworkArchitecture<T> architecture, int hiddenLayerCount, int hiddenLayerSize, int outputSize) Creates a standard feed-forward neural network with configurable hidden layers.
IEnumerable<ILayer<T>> CreateDefaultLayoutGraphLayers(int inputDim, int hiddenDim, int numGraphLayers, int numClasses, int maxNodes) Creates default LayoutGraph layers for graph-based layout analysis.
IEnumerable<ILayer<T>> CreateDefaultLayoutLMLayers(int hiddenDim, int numLayers, int numHeads, int vocabSize, int maxSequenceLength, int numClasses, int patchSize, int maxPosition2D) Creates default LayoutLM (v1) layers for document understanding with layout-aware pre-training.
IEnumerable<ILayer<T>> CreateDefaultLayoutLMv2Layers(int hiddenDim, int numLayers, int numHeads, int vocabSize, int imageSize, int visualBackboneChannels, int numClasses, int maxPosition2D) Creates default LayoutLMv2 layers for document understanding with visual features.
IEnumerable<ILayer<T>> CreateDefaultLayoutLMv3Layers(NeuralNetworkArchitecture<T> architecture, int hiddenDim, int numLayers, int numHeads, int vocabSize, int imageSize, int patchSize, int numClasses) Creates default LayoutLMv3 layers for document understanding.
IEnumerable<ILayer<T>> CreateDefaultLayoutXLMLayers(int hiddenDim, int numLayers, int numHeads, int vocabSize, int imageSize, int visualBackboneChannels, int numClasses, int maxPosition2D) Creates default LayoutXLM layers for multilingual document understanding.
IEnumerable<ILayer<T>> CreateDefaultLiLTLayers(int hiddenDim, int numLayers, int numHeads, int vocabSize, int numClasses, int maxPosition2D, int channelShrinkRatio) Creates default LiLT (Language-Independent Layout Transformer) layers.
IEnumerable<ILayer<T>> CreateDefaultLinkPredictionLayers(NeuralNetworkArchitecture<T> architecture, int hiddenDim, int embeddingDim, int numLayers, double dropoutRate) Creates default layers for a Link Prediction model encoder.
IEnumerable<ILayer<T>> CreateDefaultLiteDVDNetLayers(int inputChannels, int numFeatures, int inputBlockIntermediateChannels) Creates the LiteDVDNet denoising stack (Ilchenko & Stirenko, IJIGSP 2025).
IEnumerable<ILayer<T>> CreateDefaultMERTLayers(int hiddenDim, int numLayers, int numHeads, int feedForwardDim, double dropoutRate) Creates default layers for MERT music foundation model.
IEnumerable<ILayer<T>> CreateDefaultMGTSDLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int hiddenDimension, int numLayers, int numHeads, int numGranularities, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultMGTSDLayers(int contextLength, int forecastHorizon, int hiddenDimension, int numLayers, int numHeads, int numGranularities, double dropout) Creates default layers for the MG-TSD multi-granularity diffusion model.
IEnumerable<ILayer<T>> CreateDefaultMIAVSRLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int windowSize, int numHeads, int feedForwardRatio, int numPropagationBranches, int blocksPerBranch, int scaleFactor, int reconstructionChannels, SpyNetLayer<T> flowEstimator) Creates the MIA-VSR layers in the exact order its network binds them (Zhou et al., “Video Super-Resolution Transformer with Masked Inter&Intra-Frame Attention”, CVPR 2024).
IEnumerable<ILayer<T>> CreateDefaultMMDiTLayers(int latentChannels, int hiddenSize, int numLayers, int numHeads) Creates default MMDiT (Multi-Modal Diffusion Transformer) layers with joint attention.
IEnumerable<ILayer<T>> CreateDefaultMOIRAILayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int numFeatures, int[] patchSizes, int hiddenDim, int numLayers, int numHeads, int intermediateSize, int numMixtures, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultMOIRAILayers(int contextLength, int forecastHorizon, int numFeatures, int[] patchSizes, int hiddenDim, int numLayers, int numHeads, int intermediateSize, int numMixtures, double dropout) Creates the default layers for MOIRAI (Salesforce’s Universal Time Series Foundation Model).
IEnumerable<ILayer<T>> CreateDefaultMOMENTLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDim, int numLayers, int numHeads, int intermediateSize, double dropout, int? numClasses) Creates the default layers for MOMENT (Multi-task Optimization through Masked Encoding for Time series) foundation model.
IEnumerable<ILayer<T>> CreateDefaultMPSENetLayers(int hiddenDim, int numLayers, int numAttentionHeads, int feedForwardDim, int numFreqBins, double dropoutRate, int maxSequenceLength) Creates default layers for MP-SENet (Lu et al., 2023).
IEnumerable<ILayer<T>> CreateDefaultMQCNNLayers(NeuralNetworkArchitecture<T> architecture, int lookbackWindow, int forecastHorizon, int numFeatures, int numQuantiles, int encoderChannels, int decoderChannels, int numEncoderLayers, int numDecoderLayers, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultMQCNNLayers(int lookbackWindow, int forecastHorizon, int numFeatures, int numQuantiles, int encoderChannels, int decoderChannels, int numEncoderLayers, int numDecoderLayers, double dropout) Creates the default layer configuration for an MQCNN (Multi-Quantile CNN) model.
IEnumerable<ILayer<T>> CreateDefaultMRLLayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int maxEmbeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultMRLLayers(int vocabSize, int maxEmbeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Creates default layers for a Matryoshka Representation Learning (MRL) model.
IEnumerable<ILayer<T>> CreateDefaultMT3Layers(int numMels, int encoderDim, int numEncoderLayers, int decoderDim, int numDecoderLayers, int numAttentionHeads, int vocabSize, double dropoutRate) Creates layers for MT3 multi-track music transcription (Gardner et al., 2022).
IEnumerable<ILayer<T>> CreateDefaultMTGNNLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int forecastHorizon, int numNodes, int numFeatures, int hiddenDim, int nodeEmbeddingDim, int numLayers, int mixHopDepth) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultMTGNNLayers(int sequenceLength, int forecastHorizon, int numNodes, int numFeatures, int hiddenDim, int nodeEmbeddingDim, int numLayers, int mixHopDepth) Creates default layers for MTGNN (Multivariate Time-series Graph Neural Network).
IEnumerable<ILayer<T>> CreateDefaultMadmomBeatTrackerLayers(int numBands, int rnnHiddenSize, int numRnnLayers, double dropoutRate) Creates layers for Madmom neural beat tracker (Bock et al., 2016).
IEnumerable<ILayer<T>> CreateDefaultMamba2Layers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int numFeatures, int modelDim, int stateDim, int numHeads, int expandFactor, int numLayers, double dropout, int convKernelSize, int chunkSize) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultMamba2Layers(int contextLength, int forecastHorizon, int numFeatures, int modelDim, int stateDim, int numHeads, int expandFactor, int numLayers, double dropout, int convKernelSize, int chunkSize) Creates default layers for the Mamba-2 time series forecasting model.
IEnumerable<ILayer<T>> CreateDefaultMambaLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int numFeatures, int modelDim, int stateDim, int expandFactor, int numLayers, double dropout, int convKernelSize) Creates default layers for the Mamba (Selective State Space Model).
IEnumerable<ILayer<T>> CreateDefaultMambularLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int embeddingDimension, int stateDimension, int numLayers, int numClasses, double dropoutRate, IVectorActivationFunction<T> hiddenVectorActivation, double deltaMin, double deltaMax) Creates default layers for a Mambular (State Space Model for Tabular) model.
IEnumerable<ILayer<T>> CreateDefaultMarbleNetLayers(int numMels, int initialFilters, int numBlocks, int subBlocksPerBlock, double dropoutRate) Creates layers for MarbleNet separable convolutional VAD (NVIDIA NeMo).
IEnumerable<ILayer<T>> CreateDefaultMarketMakingLayers(NeuralNetworkArchitecture<T> architecture, int stateSize, int actionSize, int hiddenSize) Creates the default layer configuration for a Market Making agent.
IEnumerable<ILayer<T>> CreateDefaultMedGANLayers(int dataWidth, int embeddingDim, int[] autoencoderDims, int[] generatorDims, int[] discriminatorDims, bool useMinibatchAveraging, double batchNormDecay) Creates ALL layers for a medGAN medical record generator, in the fixed order the model’s reference extractor walks: autoencoder encoder, autoencoder decoder, generator, discriminator.
IEnumerable<ILayer<T>> CreateDefaultMelBandRoFormerLayers(int numMelBands, int bandEmbeddingDim, int transformerDim, int numTransformerLayers, int numAttentionHeads, int feedForwardDim, int numStems, int numFreqBins, double dropoutRate) Creates default MelBand-RoFormer layers for mel-band music source separation.
IEnumerable<ILayer<T>> CreateDefaultMelodyExtractorLayers(int numMels, int hiddenDim, int numLayers, int numPitchBins, double dropoutRate) Creates layers for neural Melody Extractor.
IEnumerable<ILayer<T>> CreateDefaultMemoryNetworkLayers(NeuralNetworkArchitecture<T> architecture, int memorySize, int embeddingSize) Creates a default Memory Network layer configuration.
IEnumerable<ILayer<T>> CreateDefaultMeshCNNLayers(NeuralNetworkArchitecture<T> architecture, int inputFeatures, int[] convChannels, int[] poolTargets, int[] fcSizes, int numNeighbors, bool useBatchNorm, double dropoutRate, bool useGlobalAveragePooling) Creates default layers for a MeshCNN architecture for mesh classification/segmentation.
IEnumerable<ILayer<T>> CreateDefaultMetaVoice1BLayers(int firstStageDim, int numFirstStageLayers, int secondStageDim, int numSecondStageLayers, int numHeads, int numCodebooks, int firstStageCodebooks, int codecLatentDim, int vocoderChannels, int vocoderUpsampleFactor, int swiGLUMultipleOf, double ropeTheta, int maxSeqLen, double dropoutRate) Paper-faithful MetaVoice-1B voice-cloning pipeline (metavoiceio/metavoice-src).
IEnumerable<ILayer<T>> CreateDefaultMiDaSLayers(int inputChannels, int inputHeight, int inputWidth, int embedDim, int numEncoderLayers) Creates default layers for MiDaS depth estimation.
IEnumerable<ILayer<T>> CreateDefaultMobileNetV2Layers(MobileNetV2Configuration configuration) Creates default layers for a MobileNetV2 network based on the specified configuration.
IEnumerable<ILayer<T>> CreateDefaultMobileNetV2Layers(NeuralNetworkArchitecture<T> architecture, MobileNetV2Configuration configuration) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultMobileNetV3Layers(MobileNetV3Configuration configuration) Creates default layers for a MobileNetV3 network based on the specified configuration.
IEnumerable<ILayer<T>> CreateDefaultMobileNetV3Layers(NeuralNetworkArchitecture<T> architecture, MobileNetV3Configuration configuration) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultMusicFlamingoLayers(int musicEncoderDim, int llmHiddenDim, int numPerceiverLayers, int numPerceiverTokens, double dropoutRate) Creates default layers for Music Flamingo music-language model.
IEnumerable<ILayer<T>> CreateDefaultMusicGenLayers(int textHiddenDim, int lmHiddenDim, int numLmLayers, int numHeads, int numCodebooks, int codebookSize, int maxTextLength, int maxAudioTokens, double dropoutRate) Creates default MusicGen layers for text-to-music generation.
IEnumerable<ILayer<T>> CreateDefaultMusicStructureAnalyzerLayers(int numMels, int hiddenDim, int numLayers, int numAttentionHeads, int numSections, double dropoutRate) Creates layers for Music Structure Analyzer.
IEnumerable<ILayer<T>> CreateDefaultMusicTaggingTransformerLayers(int numMels, int hiddenDim, int numLayers, int numAttentionHeads, int feedForwardDim, int numTags, double dropoutRate) Creates layers for Music Tagging Transformer (Won et al., 2021).
IEnumerable<ILayer<T>> CreateDefaultNBEATSLayers(NeuralNetworkArchitecture<T> architecture, int lookbackWindow, int forecastHorizon, int numStacks, int numBlocksPerStack, int hiddenSize, int numHiddenLayers) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultNBEATSLayers(int lookbackWindow, int forecastHorizon, int numStacks, int numBlocksPerStack, int hiddenSize, int numHiddenLayers) Creates layers for N-BEATS (Neural Basis Expansion Analysis for Time Series) model.
IEnumerable<ILayer<T>> CreateDefaultNHiTSLayers(NeuralNetworkArchitecture<T> architecture, int lookbackWindow, int forecastHorizon, int numStacks, int numBlocksPerStack, int hiddenSize, int numHiddenLayers, double dropout) Creates layers for N-HiTS (Neural Hierarchical Interpolation for Time Series) model.
IEnumerable<ILayer<T>> CreateDefaultNODELayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int numTrees, int treeDepth, int treeOutputDim, int outputSize, bool useBatchNorm, double dropoutRate) Creates default layers for a NODE (Neural Oblivious Decision Ensembles) model.
IEnumerable<ILayer<T>> CreateDefaultNTMLayers(NeuralNetworkArchitecture<T> architecture, int memorySize, int memoryVectorSize, int controllerSize) Creates a default configuration of layers for a Neural Turing Machine (NTM).
IEnumerable<ILayer<T>> CreateDefaultNeuralCVaRLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int sequenceLength, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultNeuralCVaRLayers(int numFeatures, int hiddenDimension, int sequenceLength, double dropoutRate) Creates default layers for a NeuralCVaR model.
IEnumerable<ILayer<T>> CreateDefaultNeuralFPLayers(int numMels, int baseFilters, int numConvBlocks, int embeddingDim, double dropoutRate) Creates default NeuralFP audio fingerprinting layers.
IEnumerable<ILayer<T>> CreateDefaultNeuralGARCHLayers(NeuralNetworkArchitecture<T> architecture, int lookbackWindow, int numAssets, int hiddenSize, int numLayers, double dropoutRate) Creates default layers for a NeuralGARCH model.
IEnumerable<ILayer<T>> CreateDefaultNeuralNetworkLayers(NeuralNetworkArchitecture<T> architecture, IReadOnlyList<int> hiddenLayerSizes) Creates a default configuration of layers for a standard neural network.
IEnumerable<ILayer<T>> CreateDefaultNeuralParametricEQLayers(int encoderDim, int numEncoderLayers, int numBands, int fftSize, double dropoutRate) Creates default layers for the Neural Parametric EQ model.
IEnumerable<ILayer<T>> CreateDefaultNeuralStressTestLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int numScenarios, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultNeuralStressTestLayers(int numFeatures, int hiddenDimension, int numScenarios, double dropoutRate) Creates default layers for a NeuralStressTest model.
IEnumerable<ILayer<T>> CreateDefaultNeuralVaRLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int sequenceLength, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultNeuralVaRLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenLayerCount, int hiddenLayerSize) Creates default layers for a NeuralVaR model.
IEnumerable<ILayer<T>> CreateDefaultNeuralVaRLayers(int numFeatures, int hiddenDimension, int sequenceLength, double dropoutRate) Creates default layers for a NeuralVaR model.
IEnumerable<ILayer<T>> CreateDefaultNodeClassificationLayers(NeuralNetworkArchitecture<T> architecture, int hiddenDim, int numLayers, double dropoutRate) Creates default layers for a Node Classification model.
IEnumerable<ILayer<T>> CreateDefaultNonStationaryTransformerLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int predictionHorizon, int numFeatures, int modelDimension, int feedForwardDim, int numEncoderLayers, int numDecoderLayers, int numHeads, int projectionDim, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultNonStationaryTransformerLayers(int sequenceLength, int predictionHorizon, int numFeatures, int modelDimension, int feedForwardDim, int numEncoderLayers, int numDecoderLayers, int numHeads, int projectionDim, double dropout) Creates default layers for a Non-stationary Transformer model.
IEnumerable<ILayer<T>> CreateDefaultNougatLayers(int hiddenDim, int numEncoderLayers, int numDecoderLayers, int numHeads, int vocabSize, int imageSize, int patchSize, int maxSequenceLength) Creates default Nougat layers for academic document understanding.
IEnumerable<ILayer<T>> CreateDefaultOccupancyLayers(NeuralNetworkArchitecture<T> architecture) Creates default layers for an occupancy detection neural network without temporal data.
IEnumerable<ILayer<T>> CreateDefaultOccupancyTemporalLayers(NeuralNetworkArchitecture<T> architecture, int historyWindowSize) Creates default layers for an occupancy detection neural network with temporal data.
IEnumerable<ILayer<T>> CreateDefaultOnsetsAndFramesLayers(int numMels, int acousticDim, int lstmHiddenSize, int numLstmLayers, int numMidiNotes, double dropoutRate) Creates default Onsets and Frames piano transcription layers.
IEnumerable<ILayer<T>> CreateDefaultOpenCLIPLayers(int visionEmbeddingDim, int textEmbeddingDim, int projectionDim, int numVisionLayers, int numTextLayers, int numVisionHeads, int numTextHeads, double dropoutRate) Creates default layers for an OpenCLIP contrastive vision-language encoder.
IEnumerable<ILayer<T>> CreateDefaultOpenSoraLayers(NeuralNetworkArchitecture<T> architecture, int height, int width, int channels, int hiddenDim, int numLayers, int numHeads) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultOpenSoraLayers(int height, int width, int channels, int hiddenDim, int numLayers, int numHeads) Creates default layers for an OpenSora video generation model.
IEnumerable<ILayer<T>> CreateDefaultOpenVoiceV2Layers(int specChannels, int hiddenDim, int numFlowBlocks, int upsampleRate, int vocabSize) OpenVoice V2 tone-color converter (Qin et al.
IEnumerable<ILayer<T>> CreateDefaultOpticalFlowLayers(int inputChannels, int inputHeight, int inputWidth, int hiddenDim) Creates layers for an optical flow estimation model (RAFT-style).
IEnumerable<ILayer<T>> CreateDefaultPANNsLayers(int numClasses, int embeddingDim, double dropoutRate) Creates the default CNN14 audio classifier layers for PANNs (Kong et al.
IEnumerable<ILayer<T>> CreateDefaultPANNsLayers(int numMels, int baseChannels, int numBlocks, int embeddingDim, int numClasses, int maxFrames, double dropoutRate, double headDropoutRate) Creates default PANNs CNN14 layers for audio classification.
IEnumerable<ILayer<T>> CreateDefaultPATEGANDiscriminatorLayers(int inputDim, int outputDim, int[] hiddenDims) Creates default PATE-GAN teacher/student discriminator layers.
IEnumerable<ILayer<T>> CreateDefaultPATEGANGeneratorLayers(int inputDim, int outputDim, int[] hiddenDims) Creates default PATE-GAN generator layers with residual connections.
IEnumerable<ILayer<T>> CreateDefaultPINNLayers(NeuralNetworkArchitecture<T> architecture, int hiddenLayerCount, int hiddenLayerSize) Creates default layers for a Physics-Informed Neural Network (PINN).
IEnumerable<ILayer<T>> CreateDefaultPSENetLayers(int imageSize, int backboneChannels, int featureChannels, int numKernels) Creates default PSENet (Progressive Scale Expansion Network) layers.
IEnumerable<ILayer<T>> CreateDefaultPatchTSTLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int predictionHorizon, int numFeatures, int patchSize, int stride, int numLayers, int numHeads, int modelDimension, int feedForwardDimension) Creates the default layers for a PatchTST (Patch Time Series Transformer) network.
IEnumerable<ILayer<T>> CreateDefaultPeakNetFPLayers(int numMels, int baseFilters, int numEncoderBlocks, int embeddingDim, double dropoutRate) Creates layers for PeakNetFP spectral peak-based fingerprinting.
IEnumerable<ILayer<T>> CreateDefaultPengiLayers(int audioEncoderDim, int llmHiddenDim, int numProjectionLayers, double dropoutRate) Creates default layers for Pengi audio language model.
IEnumerable<ILayer<T>> CreateDefaultPerceiverResamplerLayers(int visionDim, int perceiverDim, int decoderDim, int numVisionLayers, int numPerceiverLayers, int numDecoderLayers, int numLatents, int numHeads, int numPerceiverHeads, double dropoutRate) Creates default layers for perceiver-resampler VLMs (OpenFlamingo, IDEFICS, IDEFICS2, IDEFICS3).
IEnumerable<ILayer<T>> CreateDefaultPix2StructLayers(int hiddenDim, int numEncoderLayers, int numDecoderLayers, int numHeads, int vocabSize, int patchSize, int maxPatches, int maxSequenceLength) Creates default Pix2Struct layers for screenshot parsing.
IEnumerable<ILayer<T>> CreateDefaultPixelShuffleProjectorLayers(int visionDim, int shuffleDim, int decoderDim, int numVisionLayers, int numDecoderLayers, int numHeads, double dropoutRate, int patchSize) Creates layers for InternVL-style VLMs using pixel shuffle spatial compression.
IEnumerable<ILayer<T>> CreateDefaultPointCloudVLMLayers(int pointEncoderDim, int decoderDim, int numEncoderLayers, int numDecoderLayers, int numHeads, double dropoutRate) Creates layers for 3D point cloud VLMs with point-based encoder.
IEnumerable<ILayer<T>> CreateDefaultPointTransformerV3Layers(PointTransformerV3Options options, int numClasses) Creates the default PointTransformerV3 network (Wu et al., CVPR 2024; Pointcept PT-v3m1): the serialized-attention backbone with its U-Net decoder, then Pointcept’s seg_head = Linear(dec_channels[0], num_classes).
IEnumerable<ILayer<T>> CreateDefaultProprietaryAPILayers(int visionDim, int decoderDim, int numVisionLayers, int numDecoderLayers, int numHeads, double dropoutRate) Creates lightweight layers for proprietary API wrapper VLMs.
IEnumerable<ILayer<T>> CreateDefaultProprietaryASRLayers(int encoderDim, int numLayers, int numAttentionHeads, int feedForwardDim, int numMels, int vocabSize, double dropoutRate, int maxSequenceLength) Creates default layers for a proprietary API wrapper ASR model.
IEnumerable<ILayer<T>> CreateDefaultPyAnnoteLayers(int numMels, int lstmHiddenSize, int numLSTMLayers, int linearDim, int maxSpeakersPerChunk, double dropoutRate) Creates default pyannote 3.x segmentation layers.
IEnumerable<ILayer<T>> CreateDefaultQFormerGenerativeLayers(int visionDim, int qFormerDim, int decoderDim, int numVisionLayers, int numQFormerLayers, int numDecoderLayers, int numQueryTokens, int numHeads, int numQFormerHeads, double dropoutRate, int patchSize) Creates default layers for Q-Former-based generative VLMs (InstructBLIP, BLIP-3, MiniGPT-4, MiniGPT-v2).
IEnumerable<ILayer<T>> CreateDefaultQuailVadLayers(int hiddenDim, int numCNNLayers, int rnnHiddenSize, int frameSizeMs, int sampleRate) Creates default layers for Quail VAD lightweight voice activity detection.
IEnumerable<ILayer<T>> CreateDefaultQuantumNetworkLayers(NeuralNetworkArchitecture<T> architecture, int numQubits, int numQuantumLayers) Creates a default configuration of layers for a Quantum Neural Network.
IEnumerable<ILayer<T>> CreateDefaultQwen2AudioLayers(int audioEncoderDim, int numAudioEncoderLayers, int numAudioEncoderHeads, int lmHiddenDim, double dropoutRate) Creates default layers for Qwen2-Audio (audio encoder + adapter).
IEnumerable<ILayer<T>> CreateDefaultQwen2TextLayers(int vocabSize, int maxSeqLen, int hiddenSize, int numLayers, int numHeads, int numKvHeads, double ropeTheta) Paper-faithful Qwen2 text encoder stack (Yang et al.
IEnumerable<ILayer<T>> CreateDefaultRBFNetworkLayers(NeuralNetworkArchitecture<T> architecture, int hiddenSize, IRadialBasisFunction<T> rbfFunction) Creates a default Radial Basis Function (RBF) neural network layer configuration.
IEnumerable<ILayer<T>> CreateDefaultREaLTabFormerLayers(int vocabSize, int embeddingDim, int maxSeqLen, int numHeads, int numTransformerLayers, int ffnDim, double dropoutRate) Creates default autoregressive transformer layers for REaLTabFormer.
IEnumerable<ILayer<T>> CreateDefaultRNNLayers(NeuralNetworkArchitecture<T> architecture, int hiddenSize, int recurrentLayerCount) Creates a default Recurrent Neural Network (RNN) layer configuration.
IEnumerable<ILayer<T>> CreateDefaultRNNTransducerLayers(int encoderDim, int numEncoderLayers, int numEncoderHeads, int predictionDim, int numPredictionLayers, int embeddingDim, int jointDim, int numMels, int vocabSize, double dropoutRate) Creates default layers for RNN-Transducer (encoder + prediction + joint).
IEnumerable<ILayer<T>> CreateDefaultRVMLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures) Creates default layers for RVM (Robust Video Matting, Lin et al.
IEnumerable<ILayer<T>> CreateDefaultRWKV7LanguageModelLayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int modelDim, int numHeads, int numLayers, double ffnMultiplier, int maxSeqLength) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultRWKV7LanguageModelLayers(int vocabSize, int modelDim, int numHeads, int numLayers, double ffnMultiplier, int maxSeqLength) Creates default layers for the RWKV-7 “Goose” language model.
IEnumerable<ILayer<T>> CreateDefaultRWKVForecastingLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int numFeatures, int modelDim, int numHeads, int numLayers, double dropout, double globalIclrMultiplier) Creates default layers for the RWKV time series forecasting model.
IEnumerable<ILayer<T>> CreateDefaultRWKVTransducerLayers(int encoderDim, int numLayers, int numHeads, int numMels, int vocabSize, double dropoutRate, int maxSequenceLength) Creates the RWKV-Transducer encoder: conv subsampling, N RWKV time-mixing blocks, and a CTC head.
IEnumerable<ILayer<T>> CreateDefaultRealizedVolatilityTransformerLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int numAssets, int hiddenSize, int numHeads, int numLayers, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultRealizedVolatilityTransformerLayers(int sequenceLength, int numAssets, int hiddenSize, int numHeads, int numLayers, double dropoutRate) Creates default layers for a RealizedVolatilityTransformer model.
IEnumerable<ILayer<T>> CreateDefaultRelationalGCNLayers(NeuralNetworkArchitecture<T> architecture, int numNodes, int numFeatures, int numRelations, int hiddenDimension, int numLayers, int numBases, int forecastHorizon) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultRelationalGCNLayers(int numNodes, int numFeatures, int numRelations, int hiddenDimension, int numLayers, int numBases, int forecastHorizon) Creates default layers for RelationalGCN (Relational Graph Convolutional Network).
IEnumerable<ILayer<T>> CreateDefaultResNetLayers(NeuralNetworkArchitecture<T> architecture, int blockCount, int blockSize) Creates a Residual Neural Network (ResNet) with configurable blocks.
IEnumerable<ILayer<T>> CreateDefaultRoboticsActionLayers(int visionDim, int decoderDim, int actionDim, int numVisionLayers, int numDecoderLayers, int numActionLayers, int numHeads, double dropoutRate) Creates layers for robotics VLMs with action token output head.
IEnumerable<ILayer<T>> CreateDefaultRoomImpulseResponseLayers(int numEncoderBlocks, int encoderKernelSize, int encoderStride, int encoderMaxChannels, int latentDim, int numDecoderBlocks, int numNoiseBands, int noiseFilterOrder) Creates the FiNS layer stack for time-domain room impulse response estimation (Steinmetz et al., WASPAA 2021, arXiv:2107.07503).
IEnumerable<ILayer<T>> CreateDefaultS4Layers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int modelDim, int stateDim, int numLayers, bool useLowRankCorrection, int lowRankRank, int numFeatures) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultS4Layers(int contextLength, int forecastHorizon, int modelDim, int stateDim, int numLayers, bool useLowRankCorrection, int lowRankRank, int numFeatures) Creates default layers for the S4 (Structured State Space Sequence) model.
IEnumerable<ILayer<T>> CreateDefaultSAINTLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int numHeads, int numLayers, int sequenceLength, int numClasses, double dropoutRate, int? feedForwardDimension, IVectorActivationFunction<T> hiddenVectorActivation) Creates default layers for a SAINT model.
IEnumerable<ILayer<T>> CreateDefaultSALMONNLayers(int speechEncoderDim, int audioEncoderDim, int qFormerDim, int numQFormerLayers, int lmHiddenDim, double dropoutRate) Creates default layers for SALMONN (dual encoder + Q-Former).
IEnumerable<ILayer<T>> CreateDefaultSAM2Layers(int inputChannels, int inputHeight, int inputWidth, int numFeatures) Creates all SAM2 layers for backward compatibility.
IEnumerable<ILayer<T>> CreateDefaultSCNetLayers(NeuralNetworkArchitecture<T> architecture, int numClusters, int compressionDim, int numEncoderBlocks, int numDecoderBlocks, int numAttentionHeads, int feedForwardDim, int numStems, int numFreqBins, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultSCNetLayers(int numClusters, int compressionDim, int numEncoderBlocks, int numDecoderBlocks, int numAttentionHeads, int feedForwardDim, int numStems, int numFreqBins, double dropoutRate) Creates default layers for the SCNet (Sparse Compression Network) source separation model.
IEnumerable<ILayer<T>> CreateDefaultSD15UNetLayers(int latentChannels, int baseChannels, int[] channelMultipliers, int numResBlocks, int crossAttentionDim) Creates default U-Net layers for Stable Diffusion 1.5 architecture.
IEnumerable<ILayer<T>> CreateDefaultSDXLUNetLayers(int latentChannels, int baseChannels, int crossAttentionDim1, int crossAttentionDim2) Creates default U-Net layers for SDXL architecture with dual cross-attention.
IEnumerable<ILayer<T>> CreateDefaultSECBERTLayers(NeuralNetworkArchitecture<T> architecture, int maxSequenceLength, int vocabularySize, int hiddenSize, int numAttentionHeads, int numHiddenLayers, double dropoutProbability) Creates the default layer configuration for SEC-BERT.
IEnumerable<ILayer<T>> CreateDefaultSECBERTLayers(NeuralNetworkArchitecture<T> architecture, int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numClasses, double dropoutRate, string taskType) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultSECBERTLayers(int vocabularySize, int maxSequenceLength, int hiddenDimension, int numAttentionHeads, int intermediateDimension, int numLayers, int numClasses, double dropoutRate, string taskType) Creates the default layer configuration for SEC-BERT (Securities and Exchange Commission BERT).
IEnumerable<ILayer<T>> CreateDefaultSGPTLayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultSGPTLayers(int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Creates default layers for an SGPT (Sentence GPT) decoder-only embedding model.
IEnumerable<ILayer<T>> CreateDefaultSPLADELayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultSPLADELayers(int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Creates default layers for a SPLADE (Sparse Lexical and Expansion Model) embedding model.
IEnumerable<ILayer<T>> CreateDefaultSTGNNLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int forecastHorizon, int numNodes, int numFeatures, int hiddenDim, int numSpatialLayers, int numTemporalLayers) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultSTGNNLayers(int sequenceLength, int forecastHorizon, int numNodes, int numFeatures, int hiddenDim, int numSpatialLayers, int numTemporalLayers) Creates default layers for the STGNN (Spatio-Temporal Graph Neural Network) model.
IEnumerable<ILayer<T>> CreateDefaultSVTRLayers(int imageWidth, int imageHeight, int hiddenDim, int numLayers, int numHeads, int charsetSize) Creates default SVTR (Scene Text Visual Transformer Recognizer) layers.
IEnumerable<ILayer<T>> CreateDefaultSambaASRLayers(int encoderDim, int numLayers, int stateDimension, int expandFactor, int convKernelSize, int numMels, int vocabSize, double dropoutRate, int maxSequenceLength) Builds the paper-accurate SAMBA-ASR encoder (Yadav et al., “SAMBA-ASR: State-of-the-Art Speech Recognition Leveraging Structured State-Space Models”, arXiv:2501.02832, 2025).
IEnumerable<ILayer<T>> CreateDefaultScoreGradLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int forecastHorizon, int numFeatures, int hiddenDim, int numLayers) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultScoreGradLayers(int sequenceLength, int forecastHorizon, int numFeatures, int hiddenDim, int numLayers) Creates default layers for the ScoreGrad (Score-based Gradient) model.
IEnumerable<ILayer<T>> CreateDefaultSiameseLayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int embeddingDimension, int maxSequenceLength) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultSiameseLayers(int vocabSize, int embeddingDimension, int maxSequenceLength) Creates default layers for a Siamese neural network using a Transformer-based encoder.
IEnumerable<ILayer<T>> CreateDefaultSigLIP2Layers(int visionEmbeddingDim, int textEmbeddingDim, int projectionDim, int numVisionLayers, int numTextLayers, int numVisionHeads, int numTextHeads, int captioningDecoderDim, int numCaptioningDecoderLayers, int numCaptioningDecoderHeads, int mimDecoderDim, int numMimDecoderLayers, int vocabSize, bool includeCaptioningDecoder, double dropoutRate, int patchSize, int inputChannels) Creates default layers for SigLIP 2 (Multilingual Vision-Language Encoders with Improved Semantic Understanding).
IEnumerable<ILayer<T>> CreateDefaultSigLIP2TextLayers(int vocabSize, int maxSeqLen, int hiddenSize, int numLayers, int numHeads) Paper-faithful SigLIP2 text encoder stack (Tschannen et al.
IEnumerable<ILayer<T>> CreateDefaultSigLIPLayers(int visionEmbeddingDim, int textEmbeddingDim, int projectionDim, int numVisionLayers, int numTextLayers, int numVisionHeads, int numTextHeads, double dropoutRate) Creates default layers for a SigLIP contrastive vision-language encoder.
IEnumerable<ILayer<T>> CreateDefaultSigLIPTextLayers(int vocabSize, int maxSeqLen, int hiddenSize, int numLayers, int numHeads) Paper-faithful SigLIP text encoder stack (Zhai et al., ICCV 2023).
IEnumerable<ILayer<T>> CreateDefaultSignatureInformedTransformerLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int numHeads, int numAssets, int sequenceLength, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultSignatureInformedTransformerLayers(int numFeatures, int hiddenDimension, int numHeads, int numAssets, int sequenceLength, double dropoutRate) Creates default layers for a SignatureInformedTransformer (arXiv:2510.03129).
IEnumerable<ILayer<T>> CreateDefaultSimCSELayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultSimCSELayers(int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Creates default layers for a SimCSE (Simple Contrastive Learning of Sentence Embeddings) model.
IEnumerable<ILayer<T>> CreateDefaultSimMTMLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numLayers, int numHeads, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultSimMTMLayers(int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numLayers, int numHeads, double dropout) Creates default layers for the SimMTM masked modeling with similarity model.
IEnumerable<ILayer<T>> CreateDefaultSingleStreamFusionLayers(int visionDim, int textDim, int fusionDim, int numFusionLayers, int numHeads, double dropoutRate) Creates default layers for single-stream vision-language fusion models (VisualBERT, UNITER, Oscar, VinVL, ViLT).
IEnumerable<ILayer<T>> CreateDefaultSlowFastLayers(int inputChannels, int inputHeight, int inputWidth, int numClasses, int slowChannels, int fastChannels, int alpha) Creates all SlowFast layers for backward compatibility (returns only slow pathway).
IEnumerable<ILayer<T>> CreateDefaultSonataLayers(SonataOptions options, int numClasses) Creates the default Sonata network (Wu et al., CVPR 2025; Pointcept Sonata v1m1): the encoder-only PTv3 m2 backbone whose features are up-cast to full resolution, then the linear probe Linear(sum(enc_channels), num_classes) of Sonata’s…
IEnumerable<ILayer<T>> CreateDefaultSourceSeparationLayers(int numMels, int baseChannels, int numSources, int maxFrames, double dropoutRate) Creates default music source separation layers (U-Net style).
IEnumerable<ILayer<T>> CreateDefaultSpanBasedNERLayers(int hiddenDimension, int numAttentionHeads, int numTransformerLayers, int intermediateDimension, int spanEmbeddingDimension, int numLabels, double dropoutRate)
IEnumerable<ILayer<T>> CreateDefaultSpeakerEmbeddingLayers(int numMels, int hiddenDim, int embeddingDim, int numLayers, int maxFrames, double dropoutRate) Creates default speaker embedding layers for speaker verification and identification.
IEnumerable<ILayer<T>> CreateDefaultSpeakerLMLayers(int numMels, int embeddingDim, int lmHiddenDim, int numLMLayers, int numHeads, double dropoutRate) Creates default layers for SpeakerLM language-model-based speaker recognition.
IEnumerable<ILayer<T>> CreateDefaultSpikingFullSubNetLayers(int numFreqBins, int fullBandHiddenSize, int subBandHiddenSize, int fullBandLayers, int subBandLayers, double dropoutRate) Creates default layers for Spiking-FullSubNet speech enhancement.
IEnumerable<ILayer<T>> CreateDefaultSpikingLayers(NeuralNetworkArchitecture<T> architecture, SpikingNeuronType neuronType, double tau, double refractoryPeriod, bool useLayerNormalization, bool useOutputConversion, int timeSteps, IReadOnlyList<int> hiddenLayerSizes) Creates default layers for a Spiking Neural Network (SNN).
IEnumerable<ILayer<T>> CreateDefaultSpiralNetLayers(NeuralNetworkArchitecture<T> architecture, int inputFeatures, int spiralLength, int[] convChannels, double[] poolRatios, int[] fcSizes, bool useBatchNorm, double dropoutRate, bool useGlobalAveragePooling) Creates the default layer sequence for a SpiralNet mesh neural network.
IEnumerable<ILayer<T>> CreateDefaultSqueezeformerLayers(int encoderDim, int numLayers, int numAttentionHeads, int feedForwardExpansionFactor, int numMels, int vocabSize, double dropoutRate, int maxSequenceLength, bool useLayerNormalization) Creates default layers for a Squeezeformer encoder with temporal U-Net structure.
IEnumerable<ILayer<T>> CreateDefaultStableAudioLayers(int textHiddenDim, int latentDim, int ditHiddenDim, int numDitBlocks, int numHeads, int maxTextLength, int maxAudioLength, double dropoutRate) Creates default Stable Audio layers for text-to-audio generation.
IEnumerable<ILayer<T>> CreateDefaultStreamingConformerLayers(int encoderDim, int numLayers, int numAttentionHeads, int feedForwardExpansionFactor, int numMels, int vocabSize, double dropoutRate, int maxSequenceLength) Creates default layers for a streaming Conformer with chunked attention.
IEnumerable<ILayer<T>> CreateDefaultSundialLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDim, int numLayers, int numHeads, int intermediateSize, double dropout) Creates the default layers for Sundial (decoder-only foundation model).
IEnumerable<ILayer<T>> CreateDefaultT5TextLayers(int vocabSize, int hiddenSize, int numLayers, int numHeads, int numRelativePositionBuckets, int relativePositionMaxDistance) Paper-faithful T5 text encoder stack (Raffel et al., JMLR 2020).
IEnumerable<ILayer<T>> CreateDefaultTCNLayers(NeuralNetworkArchitecture<T> architecture, int lookbackWindow, int forecastHorizon, int numFeatures, int numChannels, int kernelSize, int numLayers, double dropout, bool useResidualConnections) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTCNLayers(int lookbackWindow, int forecastHorizon, int numFeatures, int numChannels, int kernelSize, int numLayers, double dropout, bool useResidualConnections) Creates the default layers for a TCN (Temporal Convolutional Network) model.
IEnumerable<ILayer<T>> CreateDefaultTESTLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int textEmbeddingDimension, int numLayers, int numHeads, int numPrototypes, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTESTLayers(int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int textEmbeddingDimension, int numLayers, int numHeads, int numPrototypes, double dropout) Creates default layers for the TEST model (text-aligned time series embeddings).
IEnumerable<ILayer<T>> CreateDefaultTFCLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int hiddenDimension, int projectionDimension, int numTimeLayers, int numFreqLayers, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTFCLayers(int contextLength, int forecastHorizon, int hiddenDimension, int projectionDimension, int numTimeLayers, int numFreqLayers, double dropout) Creates default layers for the TF-C time-frequency consistency model.
IEnumerable<ILayer<T>> CreateDefaultTFGridNetLayers(int numFreqBins, int hiddenDim, int embeddingDim, int numBlocks, int numAttentionHeads, double dropoutRate) Creates default TF-GridNet layers for time-frequency grid speech enhancement.
IEnumerable<ILayer<T>> CreateDefaultTFTLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int predictionHorizon, int numFeatures, int hiddenSize, int numHeads, int numLayers, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTFTLayers(int sequenceLength, int predictionHorizon, int numFeatures, int hiddenSize, int numHeads, int numLayers, double dropout) Creates the default layers for a Temporal Fusion Transformer (TFT) architecture.
IEnumerable<ILayer<T>> CreateDefaultTOTEMLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int hiddenDimension, int numLayers, int numHeads, int codebookDimension, double dropout) Creates default layers for the TOTEM VQ-VAE tokenization model.
IEnumerable<ILayer<T>> CreateDefaultTOTOLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numLayers, int numHeads, int intermediateSize, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTOTOLayers(int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numLayers, int numHeads, int intermediateSize, double dropout) Creates default layers for the TOTO observability foundation model.
IEnumerable<ILayer<T>> CreateDefaultTRIELayers(int imageSize, int visualDim, int textDim, int graphDim, int numEntityTypes, int maxEntities) Creates default TRIE (Text Reading and Information Extraction) layers.
IEnumerable<ILayer<T>> CreateDefaultTS2VecLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int hiddenDimension, int outputDimension, int numLayers, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTS2VecLayers(int contextLength, int forecastHorizon, int hiddenDimension, int outputDimension, int numLayers, double dropout) Creates default layers for the TS2Vec contrastive learning model.
IEnumerable<ILayer<T>> CreateDefaultTSDiffLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int forecastHorizon, int hiddenDimension, int numResidualBlocks, int numAttentionHeads, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTSDiffLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int numFeatures, int hiddenDim, int numResidualBlocks, int numAttentionHeads) Creates default layers for the TSDiff self-guided diffusion model.
IEnumerable<ILayer<T>> CreateDefaultTSDiffLayers(int sequenceLength, int forecastHorizon, int hiddenDimension, int numResidualBlocks, int numAttentionHeads, double dropout) Creates default layers for the TSDiff self-guided diffusion model.
IEnumerable<ILayer<T>> CreateDefaultTSDiffLayers(int sequenceLength, int numFeatures, int hiddenDim, int numResidualBlocks, int numAttentionHeads) Creates the default layer configuration for TSDiff (Time Series Diffusion).
IEnumerable<ILayer<T>> CreateDefaultTSMixerLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int predictionHorizon, int numFeatures, int hiddenDim, int numBlocks, double feedForwardExpansion, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTSMixerLayers(int sequenceLength, int predictionHorizon, int numFeatures, int hiddenDim, int numBlocks, double feedForwardExpansion, double dropout) Creates default layers for a TSMixer model.
IEnumerable<ILayer<T>> CreateDefaultTVAEDecoderLayers(int latentDim, int outputDim, int[] hiddenDims) Creates default TVAE decoder layers: hidden layers with ReLU, final identity output layer.
IEnumerable<ILayer<T>> CreateDefaultTVAEEncoderLayers(int inputDim, int latentDim, int[] hiddenDims) Creates default TVAE encoder layers: hidden layers with ReLU, final layer outputs 2*latentDim (mean and logvar concatenated).
IEnumerable<ILayer<T>> CreateDefaultTabDDPMDenoiserLayers(int inputDim, int[] hiddenDims, double dropoutRate) Creates default TabDDPM denoiser MLP layers: SiLU hidden layers with optional dropout.
IEnumerable<ILayer<T>> CreateDefaultTabDPTLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int embeddingDimension, int numHeads, int numLayers, int numClasses, double dropoutRate, int? feedForwardDimension, IVectorActivationFunction<T> hiddenVectorActivation, bool useLayerNorm) Creates default layers for a TabDPT (Tabular Data Pre-Training) model.
IEnumerable<ILayer<T>> CreateDefaultTabFlowTimeProjectionLayers(int timeEmbDim) Creates default time projection layers for a TabFlow generator.
IEnumerable<ILayer<T>> CreateDefaultTabFlowVelocityLayers(int inputDim, int outputDim, int[] hiddenDims, double dropoutRate) Creates default velocity MLP layers for a TabFlow generator.
IEnumerable<ILayer<T>> CreateDefaultTabMLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int[] hiddenDimensions, int numClasses, double dropoutRate) Creates default layers for a TabM (Parameter-Efficient Ensemble) model.
IEnumerable<ILayer<T>> CreateDefaultTabNetLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int numSteps, int numClasses, double dropoutRate) Creates default layers for a TabNet model.
IEnumerable<ILayer<T>> CreateDefaultTabPFNLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int embeddingDimension, int numHeads, int numLayers, int numClasses, double dropoutRate, int? feedForwardDimension, IVectorActivationFunction<T> hiddenVectorActivation) Creates default layers for a TabPFN (Prior-Fitted Network) model.
IEnumerable<ILayer<T>> CreateDefaultTabRLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int embeddingDimension, int numLayers, int numClasses, double dropoutRate) Creates default layers for a TabR (Retrieval-Augmented) model.
IEnumerable<ILayer<T>> CreateDefaultTabSynDecoderLayers(int latentDim, int outputDim, int[] hiddenDims) Creates default VAE decoder layers for TabSyn latent diffusion.
IEnumerable<ILayer<T>> CreateDefaultTabSynDiffusionLayers(int inputDim, int outputDim, int[] hiddenDims) Creates default latent diffusion denoiser layers for TabSyn.
IEnumerable<ILayer<T>> CreateDefaultTabSynEncoderLayers(int inputDim, int latentDim, int[] hiddenDims) Creates default VAE encoder layers for TabSyn latent diffusion.
IEnumerable<ILayer<T>> CreateDefaultTabTransformerLayers(NeuralNetworkArchitecture<T> architecture, int numFeatures, int hiddenDimension, int numHeads, int numLayers, int sequenceLength, int numClasses, double dropoutRate, int? feedForwardDimension) Creates default layers for a TabTransformer model.
IEnumerable<ILayer<T>> CreateDefaultTableTransformerLayers(int imageSize, int hiddenDim, int numEncoderLayers, int numDecoderLayers, int numHeads, int numQueries, int numStructureClasses) Creates default layers for TableTransformer model.
IEnumerable<ILayer<T>> CreateDefaultTabularClassifierLayers(int inputDim, int outputDim, int[] hiddenDims) Creates default classifier MLP layers for TableGAN and CTAB-GAN+ auxiliary classification.
IEnumerable<ILayer<T>> CreateDefaultTabularGANDiscriminatorLayers(int inputDim, int outputDim, int[] hiddenDims, double dropoutRate) Creates default discriminator MLP layers for GAN-based tabular generators.
IEnumerable<ILayer<T>> CreateDefaultTabularGANGeneratorLayers(int inputDim, int outputDim, int[] hiddenDims, bool useBatchNorm, double dropoutRate) Creates default generator MLP layers for GAN-based tabular generators (CopulaGAN, DP-CTGAN, CTAB-GAN+, TableGAN).
IEnumerable<ILayer<T>> CreateDefaultTempogramLayers(int fftSize, int onsetHiddenDim, int numOnsetLayers, int numTempoBins, double dropoutRate) Creates layers for Tempogram neural tempo estimation.
IEnumerable<ILayer<T>> CreateDefaultTemporalGCNLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int forecastHorizon, int numNodes, int numFeatures, int hiddenDim, int numGCNLayers, int numTemporalLayers) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTemporalGCNLayers(int sequenceLength, int forecastHorizon, int numNodes, int numFeatures, int hiddenDim, int numGCNLayers, int numTemporalLayers) Creates default layers for TemporalGCN (Temporal Graph Convolutional Network).
IEnumerable<ILayer<T>> CreateDefaultTimeBridgeLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numLayers, int numHeads, int intermediateSize, int bridgeDimension, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTimeBridgeLayers(int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numLayers, int numHeads, int intermediateSize, int bridgeDimension, double dropout) Creates default layers for the TimeBridge non-stationarity foundation model.
IEnumerable<ILayer<T>> CreateDefaultTimeDiffLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int hiddenDimension, int numLayers, int numHeads, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTimeDiffLayers(int contextLength, int forecastHorizon, int hiddenDimension, int numLayers, int numHeads, double dropout) Creates default layers for the TimeDiff non-autoregressive diffusion model.
IEnumerable<ILayer<T>> CreateDefaultTimeGANEmbedderLayers(int inputDim, int outputDim, int[] hiddenDims) Creates default embedder layers for TimeGAN’s embedding network.
IEnumerable<ILayer<T>> CreateDefaultTimeGANRecoveryLayers(int inputDim, int outputDim, int[] hiddenDims) Creates default recovery layers for TimeGAN’s recovery network.
IEnumerable<ILayer<T>> CreateDefaultTimeGANSupervisorLayers(int embeddingDim, int[] hiddenDims) Creates default supervisor layers for TimeGAN’s supervisor network.
IEnumerable<ILayer<T>> CreateDefaultTimeGPTLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int numFeatures, int hiddenDim, int numLayers, int numHeads, double dropout) Creates default layers for the TimeGPT-style model.
IEnumerable<ILayer<T>> CreateDefaultTimeLLMLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int numFeatures, int patchLength, int patchStride, int llmDim, int numPrototypes, int numLayers, int numHeads, double dropout) Creates the default layers for Time-LLM (LLM Reprogramming for Time Series).
IEnumerable<ILayer<T>> CreateDefaultTimeMAELayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numEncoderLayers, int numDecoderLayers, int numHeads, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTimeMAELayers(int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numEncoderLayers, int numDecoderLayers, int numHeads, double dropout) Creates default layers for the TimeMAE masked autoencoder model.
IEnumerable<ILayer<T>> CreateDefaultTimeMachineLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int modelDim, int stateDim, int expandFactor, int convKernelSize, int numFeatures, double dropoutRate) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTimeMachineLayers(int contextLength, int forecastHorizon, int modelDim, int stateDim, int expandFactor, int convKernelSize, int numFeatures, double dropoutRate) Creates default layers for the TimeMachine (Time Series State Space Model) architecture.
IEnumerable<ILayer<T>> CreateDefaultTimeMoELayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDim, int numLayers, int numHeads, int intermediateSize, int numExperts, int numActiveExperts, double dropout) Creates the default layers for Time-MoE (Mixture of Experts foundation model).
IEnumerable<ILayer<T>> CreateDefaultTimeSformerLayers(int inputChannels, int inputHeight, int inputWidth, int embedDim, int numHeads, int numLayers, int numFrames, int patchSize, int numClasses, bool useJointSpaceTimeAttention) Creates default layers for TimeSformer video classification.
IEnumerable<ILayer<T>> CreateDefaultTimerLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int numFeatures, int patchLength, int patchStride, int hiddenDim, int numLayers, int numHeads, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTimerLayers(int contextLength, int forecastHorizon, int numFeatures, int patchLength, int patchStride, int hiddenDim, int numLayers, int numHeads, double dropout) Creates default layers for the Timer (Generative Pre-Training) model.
IEnumerable<ILayer<T>> CreateDefaultTimesFMLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int numFeatures, int patchLength, int hiddenDim, int numLayers, int numHeads, double dropout, int outputPatchLength) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTimesFMLayers(int contextLength, int forecastHorizon, int numFeatures, int patchLength, int hiddenDim, int numLayers, int numHeads, double dropout, int outputPatchLength) Creates the default layer configuration for a TimesFM (Time Series Foundation Model).
IEnumerable<ILayer<T>> CreateDefaultTimesNetLayers(NeuralNetworkArchitecture<T> architecture, int sequenceLength, int predictionHorizon, int numFeatures, int modelDimension, int feedForwardDim, int numLayers, int topK, int convKernelSize, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTimesNetLayers(int sequenceLength, int predictionHorizon, int numFeatures, int modelDimension, int feedForwardDim, int numLayers, int topK, int convKernelSize, double dropout) Creates default layers for the TimesNet (Temporal 2D-Variation Modeling) architecture.
IEnumerable<ILayer<T>> CreateDefaultTinyTimeMixersLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int numFeatures, int patchLength, int hiddenDim, int numMixerLayers, int expansionFactor, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultTinyTimeMixersLayers(int contextLength, int forecastHorizon, int numFeatures, int patchLength, int hiddenDim, int numMixerLayers, int expansionFactor, double dropout) Creates the default layers for Tiny Time Mixers (TTM) foundation model.
IEnumerable<ILayer<T>> CreateDefaultTitaNetLayers(int numMels, int encoderDim, int embeddingDim, int numBlocks, int attentivePoolingDim, double dropoutRate) Creates default TitaNet speaker embedding layers.
IEnumerable<ILayer<T>> CreateDefaultTokenReductionVLMLayers(int visionDim, int reducedDim, int decoderDim, int numVisionLayers, int numDecoderLayers, int numHeads, double dropoutRate, int patchSize) Creates layers for VLMs with semantic-aware token reduction/downsampling.
IEnumerable<ILayer<T>> CreateDefaultTransformerLayers(TransformerArchitecture<T> architecture) Creates a default Transformer neural network with pre-configured encoder and decoder layers.
IEnumerable<ILayer<T>> CreateDefaultTransformerNERLayers(int hiddenDimension, int numAttentionHeads, int numTransformerLayers, int intermediateDimension, int numLabels, int maxSequenceLength, double dropoutRate, bool useCRF) Creates the default layer stack for a transformer-based NER model (BERT-NER, RoBERTa-NER, etc.).
IEnumerable<ILayer<T>> CreateDefaultUNet3DLayers(NeuralNetworkArchitecture<T> architecture, int voxelResolution, int numEncoderBlocks, int baseFilters) Creates default layers for a 3D U-Net architecture for volumetric segmentation.
IEnumerable<ILayer<T>> CreateDefaultUPRNetLayers(NeuralNetworkArchitecture<T> architecture) Creates the 47-layer native UPR-Net graph from the authors’ released base model.
IEnumerable<ILayer<T>> CreateDefaultUSMConformerLayers(int encoderDim, int numLayers, int numAttentionHeads, int feedForwardExpansionFactor, int convKernelSize, int numMels, int vocabSize, double dropoutRate) Builds a USM / Chirp encoder from GENUINE Conformer blocks (Zhang et al.
IEnumerable<ILayer<T>> CreateDefaultUniTSLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int hiddenDim, int numLayers, int numHeads, int[] convKernelSizes, double dropout, string taskType, int numClasses) Creates default layers for the UniTS (Unified Time Series) model.
IEnumerable<ILayer<T>> CreateDefaultUnifiedBidirectionalLayers(int visionDim, int sharedDim, int understandingDim, int generationDim, int numEncoderLayers, int numUnderstandingLayers, int numGenerationLayers, int numHeads, double dropoutRate) Creates layers for unified bidirectional VLMs with dual decoder heads.
IEnumerable<ILayer<T>> CreateDefaultUnifiedGenerationLayers(int visionDim, int decoderDim, int regressionDim, int numVisionLayers, int numDecoderLayers, int numRegressionLayers, int numHeads, double dropoutRate) Creates default layers for unified generation VLMs (Emu, Emu2, Emu3).
IEnumerable<ILayer<T>> CreateDefaultUnifiedMultimodalLayers(NeuralNetworkArchitecture<T> architecture, int embeddingDimension, int numTransformerLayers) Creates default layers for the UnifiedMultimodalNetwork.
IEnumerable<ILayer<T>> CreateDefaultUniversalDELayers(NeuralNetworkArchitecture<T> architecture, int hiddenLayerCount, int hiddenLayerSize) Creates default layers for a Universal Differential Equation (UDE) network.
IEnumerable<ILayer<T>> CreateDefaultVAEDecoderLayers(int latentChannels, int outputChannels, int baseChannels) Creates default VAE decoder layers for latent diffusion models.
IEnumerable<ILayer<T>> CreateDefaultVAEEncoderLayers(int inputChannels, int latentChannels, int baseChannels) Creates default VAE encoder layers for latent diffusion models.
IEnumerable<ILayer<T>> CreateDefaultVAELayers(NeuralNetworkArchitecture<T> architecture, int latentSize) Creates a default Variational Autoencoder (VAE) with pre-configured layers.
IEnumerable<ILayer<T>> CreateDefaultVALLELayers(int arHiddenDim, int numARLayers, int numARHeads, int codebookSize, double dropoutRate) Creates default layers for VALL-E AR stage.
IEnumerable<ILayer<T>> CreateDefaultVGGLayers(NeuralNetworkArchitecture<T> architecture, VGGConfiguration configuration) Creates layers for a VGG network based on the specified configuration.
IEnumerable<ILayer<T>> CreateDefaultVITSLayers(int encoderDim, int hiddenDim, int decoderDim, int numEncoderLayers, int numFlowLayers, int numDecoderLayers, int numHeads, double dropoutRate, int inputFeatures) Creates default layers for VITS-family end-to-end TTS (VITS, VITS2, YourTTS, Piper, MeloTTS).
IEnumerable<ILayer<T>> CreateDefaultVRTLayers(int inputChannels, int inputHeight, int inputWidth, int embedDim, int numFrames, int numBlocks, int scaleFactor) Creates layers for a VRT (Video Restoration Transformer) model.
IEnumerable<ILayer<T>> CreateDefaultVariationalPINNLayers(NeuralNetworkArchitecture<T> architecture, int hiddenLayerCount, int hiddenLayerSize) Creates default layers for a Variational Physics-Informed Neural Network (VPINN).
IEnumerable<ILayer<T>> CreateDefaultViTLayers(int embeddingDim, int numLayers, int numHeads, double dropoutRate, int patchSize) Creates default layers for a standard Vision Transformer (ViT) encoder.
IEnumerable<ILayer<T>> CreateDefaultVideoDenoisingLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures, int temporalFrames)
IEnumerable<ILayer<T>> CreateDefaultVideoInpaintingLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures) Creates layers for a video inpainting model (encoder-transformer-decoder).
IEnumerable<ILayer<T>> CreateDefaultVideoLinearProjectorVLMLayers(int visionDim, int decoderDim, int numVisionLayers, int numDecoderLayers, int numHeads, double dropoutRate, int imageHeight, int imageWidth, int imageChannels, int patchSize) Builds a video VLM with a plain LINEAR projector and NO dedicated temporal-aggregation transformer — the paper-faithful architecture for Video-LLaVA (Lin et al.
IEnumerable<ILayer<T>> CreateDefaultVideoMAELayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures, int numClasses, int tubeletSize) Creates default layers for VideoMAE (Video Masked Autoencoder) action recognition model.
IEnumerable<ILayer<T>> CreateDefaultVideoPoolingVLMLayers(int visionDim, int decoderDim, int numVisionLayers, int numDecoderLayers, int numHeads, double dropoutRate, int imageHeight, int imageWidth, int imageChannels, int patchSize, int poolFactor) Builds a video VLM whose temporal aggregation is parameter-free adaptive SPATIAL POOLING of the per-frame ViT features — the paper-faithful mechanism for PLLaVA (Xu et al.
IEnumerable<ILayer<T>> CreateDefaultVideoQFormerVLMLayers(int visionDim, int qFormerDim, int decoderDim, int numVisionLayers, int numQFormerLayers, int numDecoderLayers, int numQueryTokens, int numHeads, int numQFormerHeads, double dropoutRate, int imageHeight, int imageWidth, int imageChannels, int patchSize) Builds a video VLM whose temporal aggregation is a Q-Former — a stack of learnable query tokens that cross-attend to the per-frame ViT features and resample them to a fixed token set — the paper-faithful mechanism for VideoChat2 (Li et al.
IEnumerable<ILayer<T>> CreateDefaultVideoSTCVLMLayers(int visionDim, int decoderDim, int numVisionLayers, int numDecoderLayers, int visionNumHeads, int decoderNumHeads, int decoderNumKeyValueHeads, int visionFfnDim, int decoderFfnDim, double dropoutRate, int imageHeight, int imageWidth, int imageChannels, int patchSize, int maxSequenceLength, double ropeTheta, bool enableSpatialTemporalConv, int stcKernelSize, int stcStride, int stcPadding, int stcStageDepth, int stcMlpDepth) Builds the VideoLLaMA 2 default architecture: CLIP-style residual ViT, the complete STCConnectorLayer, and a Mistral-style RMSNorm/RoPE/GQA decoder with a gated SwiGLU feed-forward network.
IEnumerable<ILayer<T>> CreateDefaultVideoStabilizationLayers(int inputChannels, int inputHeight, int inputWidth) Creates layers for a video stabilization model (StabNet-style).
IEnumerable<ILayer<T>> CreateDefaultVideoSuperResolutionLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures, int numResBlocks, int scaleFactor, bool useTemporalConsistency) Creates layers for a video super-resolution model (Real-ESRGAN/BasicVSR++ style).
IEnumerable<ILayer<T>> CreateDefaultVideoTemporalVLMLayers(int visionDim, int temporalDim, int decoderDim, int numVisionLayers, int numTemporalLayers, int numDecoderLayers, int numHeads, double dropoutRate, int imageHeight, int imageWidth, int imageChannels, int patchSize) Creates layers for video VLMs with temporal aggregation across frames.
IEnumerable<ILayer<T>> CreateDefaultVideoUNetLayers(int latentChannels, int baseChannels, int crossAttentionDim, int numTemporalLayers) Creates default video U-Net layers with temporal attention for video diffusion models.
IEnumerable<ILayer<T>> CreateDefaultVisionAdapterLayers(int visionDim, int adapterDim, int decoderDim, int numVisionLayers, int numDecoderLayers, int numHeads, double dropoutRate, int patchSize) Creates layers for VLMs with vision adapter and learnable gating.
IEnumerable<ILayer<T>> CreateDefaultVisionTSLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDim, int numLayers, int numHeads, int intermediateSize, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultVisionTSLayers(int contextLength, int forecastHorizon, int patchLength, int hiddenDim, int numLayers, int numHeads, int intermediateSize, double dropout) Creates the default layers for VisionTS (Visual MAE for time series).
IEnumerable<ILayer<T>> CreateDefaultVisualExpertVLMLayers(int visionDim, int decoderDim, int visualExpertDim, int numVisionLayers, int numDecoderLayers, int numHeads, int numVisualExpertHeads, double dropoutRate) Creates default layers for visual expert VLM architecture (CogVLM pattern).
IEnumerable<ILayer<T>> CreateDefaultVoiceCloningLayers(int encoderDim, int speakerDim, int decoderDim, int numEncoderLayers, int numDecoderLayers, int numHeads, double dropoutRate) Creates default layers for voice cloning TTS (OpenVoice, MetaVoice, XTTS, Chatterbox).
IEnumerable<ILayer<T>> CreateDefaultVoiceCraftLayers(NeuralNetworkArchitecture<T> architecture, int hiddenDim, int numLayers, int numHeads, int codebookSize, double dropoutRate) Creates the default layer stack for a VoiceCraft codec language model.
IEnumerable<ILayer<T>> CreateDefaultVoxLingua107Layers(NeuralNetworkArchitecture<T> architecture, int numMels, int tdnnChannels, int embeddingDimension, int[] dilations) Creates default VoxLingua107 layers for 107-language identification.
IEnumerable<ILayer<T>> CreateDefaultVoxelCNNLayers(NeuralNetworkArchitecture<T> architecture, int voxelResolution, int numConvBlocks, int baseFilters) Creates default layers for a Voxel-based 3D Convolutional Neural Network.
IEnumerable<ILayer<T>> CreateDefaultWGANGPCriticLayers(InputType inputType) Creates the default WGAN-GP critic: LeakyReLU feature extraction followed by one unrestricted scalar score.
IEnumerable<ILayer<T>> CreateDefaultWav2SmallLayers(int numMels, int hiddenDim, int numLayers, int numAttentionHeads, int feedForwardDim, int featureEncoderDim, int numClasses, double dropoutRate) Creates default layers for the Wav2Small lightweight SER model.
IEnumerable<ILayer<T>> CreateDefaultWavLMSERLayers(int hiddenDim, int numLayers, int numAttentionHeads, int feedForwardDim, int featureEncoderDim, int numClasses, double dropoutRate) Creates default layers for the WavLM-SER emotion recognition model.
IEnumerable<ILayer<T>> CreateDefaultWavLMSpeakerLayers(int hiddenDim, int numLayers, int numAttentionHeads, int feedForwardDim, int featureEncoderDim, int embeddingDim, double dropoutRate) Creates default layers for the WavLM Speaker verification model.
IEnumerable<ILayer<T>> CreateDefaultWaveNetLayers(NeuralNetworkArchitecture<T> architecture, int lookbackWindow, int forecastHorizon, int numFeatures, int residualChannels, int skipChannels, int dilationDepth, int numStacks, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultWaveNetLayers(int lookbackWindow, int forecastHorizon, int numFeatures, int residualChannels, int skipChannels, int dilationDepth, int numStacks, double dropout) Creates the default layers for a WaveNet model adapted for time series forecasting.
IEnumerable<ILayer<T>> CreateDefaultWebRTCVadLayers(int frameSize, int hiddenDim, int numLayers, double dropoutRate) Creates layers for WebRTC neural VAD.
IEnumerable<ILayer<T>> CreateDefaultWhisperEncoderDecoderLayers(int encoderDim, int decoderDim, int numEncoderLayers, int numDecoderLayers, int numAttentionHeads, int feedForwardDim, int numMels, int vocabSize, double dropoutRate, int maxSequenceLength) Creates default layers for a Whisper-style encoder-decoder ASR.
IEnumerable<ILayer<T>> CreateDefaultWhisperLayers(int modelDim, int numEncoderLayers, int numDecoderLayers, int numHeads, int ffDim, int numMels, int maxFrames, int maxTokens, int vocabSize, double dropoutRate) Creates default Whisper layers for automatic speech recognition.
IEnumerable<ILayer<T>> CreateDefaultWhisperLayers(int numMels, int modelDimension, int numEncoderLayers, int numDecoderLayers, int numHeads, int feedForwardDim, int vocabularySize, int maxSequenceLength, double dropoutRate) Creates default layers for Whisper-style speech recognition models.
IEnumerable<ILayer<T>> CreateDefaultWord2VecLayers(NeuralNetworkArchitecture<T> architecture, int vocabSize, int embeddingDimension) Creates default layers for a Word2Vec model (Skip-Gram or CBOW).
IEnumerable<ILayer<T>> CreateDefaultXMemLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures) Creates layers for an XMem long-term video object segmentation model.
IEnumerable<ILayer<T>> CreateDefaultYingLongLayers(NeuralNetworkArchitecture<T> architecture, int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numLayers, int numHeads, int intermediateSize, double dropout) Builds the same layers as the overload without architecture, which this factory never read: the layers are sized from the other arguments (#2147).
IEnumerable<ILayer<T>> CreateDefaultYingLongLayers(int contextLength, int forecastHorizon, int patchLength, int hiddenDimension, int numLayers, int numHeads, int intermediateSize, double dropout) Creates default layers for the YingLong enterprise foundation model.
IEnumerable<ILayer<T>> CreateDefaultYuELayers(int semanticDim, int numSemanticLayers, int numSemanticHeads, int lyricsVocabSize, double dropoutRate) Creates default layers for YuE semantic stage.
IEnumerable<ILayer<T>> CreateDefaultZipformerLayers(int[] encoderDims, int[] numLayersPerStack, int[] numHeadsPerStack, int numMels, int vocabSize, double dropoutRate) Creates default layers for Zipformer (U-Net-style multi-resolution encoder).
IEnumerable<ILayer<T>> CreateDiTNoisePredictorLayers(int inputChannels, int hiddenSize, int numLayers, int numHeads, int patchSize, int contextDim, double mlpRatio) Creates layers for the DiT (Diffusion Transformer) noise predictor.
IEnumerable<ILayer<T>> CreateDiffCutDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the decoder layers for DiffCut.
IEnumerable<ILayer<T>> CreateDiffCutEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates the DiffCut encoder layers using diffusion UNet features.
IEnumerable<ILayer<T>> CreateDiffCutSegmentationDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the DiffCutSegmentation model.
IEnumerable<ILayer<T>> CreateDiffCutSegmentationEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the DiffCutSegmentation model.
IEnumerable<ILayer<T>> CreateDiffSegDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the decoder layers for DiffSeg.
IEnumerable<ILayer<T>> CreateDiffSegEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates the DiffSeg encoder layers using diffusion self-attention features.
IEnumerable<ILayer<T>> CreateDiffWaveLayers(int residualChannels, int residualLayers, int dilationCycle, int embeddingDim) Creates layers for the DiffWave audio diffusion network.
IEnumerable<ILayer<T>> CreateDreamFusionNeRFLayers(int hiddenDim, int numLayers, int posFrequencies, int dirFrequencies) Creates layers for the DreamFusion NeRF network.
IEnumerable<ILayer<T>> CreateE2FGVILayers(int channels, int height, int width, int numFeatures, int numTransformerBlocks) Creates layers for the E2FGVI video inpainting model.
IEnumerable<ILayer<T>> CreateEagleLayers(int vocabSize, int modelDimension, int numLayers, int numHeads, int maxSeqLength) Creates layers for the Eagle (RWKV-5) language model architecture.
IEnumerable<ILayer<T>> CreateEdgeSAMDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the EdgeSAM model.
IEnumerable<ILayer<T>> CreateEdgeSAMEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the EdgeSAM model.
IEnumerable<ILayer<T>> CreateEfficientSAMDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the EfficientSAM model.
IEnumerable<ILayer<T>> CreateEfficientSAMEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the EfficientSAM model.
IEnumerable<ILayer<T>> CreateEoMTDecoderLayers(int embedDim, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the lightweight mask head layers for EoMT.
IEnumerable<ILayer<T>> CreateEoMTEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int embedDim, int[] depths, double dropRate) Creates the DINOv2 ViT encoder layers for EoMT (encoder-only architecture).
IEnumerable<ILayer<T>> CreateFILMLayers(int channels, int height, int width, int numScales, int numFeatures) Creates layers for the FILM frame interpolation model.
IEnumerable<ILayer<T>> CreateFalconMambaLayers(int vocabSize, int modelDimension, int numLayers, int stateDimension, int expandFactor, int maxSeqLength) Creates layers for the FalconMamba language model architecture.
IEnumerable<ILayer<T>> CreateFastSAMDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the FastSAM model.
IEnumerable<ILayer<T>> CreateFastSAMEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the FastSAM model.
IEnumerable<ILayer<T>> CreateFinchLayers(int vocabSize, int modelDimension, int numLayers, int numHeads, int maxSeqLength) Creates layers for the Finch (RWKV-6) language model architecture.
IEnumerable<ILayer<T>> CreateFlamingoLayers(int imageSize, int channels, int patchSize, int visionHiddenDim, int lmHiddenDim, int numVisionLayers, int numPerceiverLayers, int numPerceiverTokens, int numLmLayers, int numHeads, int vocabularySize, int maxSequenceLength) Creates layers for the Flamingo native mode.
IEnumerable<ILayer<T>> CreateGLALayers(int vocabSize, int modelDimension, int numLayers, int numHeads, int maxSeqLength) Creates layers for the GLA (Gated Linear Attention) language model architecture.
IEnumerable<ILayer<T>> CreateGMFlowLayers(int channels, int height, int width, int numFeatures, int numTransformerLayers) Creates layers for the GMFlow optical flow model.
IEnumerable<ILayer<T>> CreateGatedDeltaNetLayers(int vocabSize, int modelDimension, int numLayers, int numHeads, int maxSeqLength) Creates layers for the GatedDeltaNet language model architecture.
IEnumerable<ILayer<T>> CreateGpt4VisionLayers(int imageSize, int patchSize, int visionEmbeddingDim, int embeddingDimension, int hiddenDim, int numVisionLayers, int numLanguageLayers, int numHeads, int vocabularySize) Creates layers for the GPT-4 Vision native mode.
IEnumerable<ILayer<T>> CreateGriffinLayers(int vocabSize, int modelDimension, int numLayers, int maxSeqLength, int recurrenceDimension) Creates layers for the Griffin language model architecture.
IEnumerable<ILayer<T>> CreateHawkLayers(int vocabSize, int modelDimension, int numLayers, int maxSeqLength, int recurrenceDimension) Creates layers for the Hawk language model architecture.
IEnumerable<ILayer<T>> CreateHopeNetworkLayers(int hiddenDim, int numCMSLevels, int numRecurrentLayers, int inContextLearningLevels) Creates layers for the HopeNetwork (self-modifying recurrent neural network).
IEnumerable<ILayer<T>> CreateIPAdapterLayers(int imageSize, int patchSize, int embedDim, int numLayers, int numHeads, int numTokens) Creates layers for the IP-Adapter image encoder.
IEnumerable<ILayer<T>> CreateImageBindLayers(int imageSize, int channels, int patchSize, int hiddenDim, int embeddingDimension, int numEncoderLayers, int numHeads, int vocabularySize, int maxSequenceLength, int audioSampleRate, int audioMaxDuration, int imuTimesteps, int numVideoFrames) Creates layers for the ImageBind native mode.
IEnumerable<ILayer<T>> CreateInstantNGPLayers(int numLevels, int featuresPerLevel, int mlpNumLayers, int mlpHiddenDim, int featureDim, int colorHiddenDim, int colorNumLayers) Creates layers for the Instant-NGP (Neural Graphics Primitives) model.
IEnumerable<ILayer<T>> CreateInternImageDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the UPerNet decoder layers for InternImage.
IEnumerable<ILayer<T>> CreateInternImageEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates the InternImage encoder layers using deformable convolution approximation.
IEnumerable<ILayer<T>> CreateJambaLayers(int vocabSize, int modelDimension, int numLayers, int stateDimension, int attentionInterval, int maxSeqLength) Creates layers for the Jamba hybrid language model architecture.
IEnumerable<ILayer<T>> CreateKMaXDeepLabDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the KMaXDeepLab model.
IEnumerable<ILayer<T>> CreateKMaXDeepLabEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the KMaXDeepLab model.
IEnumerable<ILayer<T>> CreateLISADecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the LISA model.
IEnumerable<ILayer<T>> CreateLISAEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the LISA model.
IEnumerable<ILayer<T>> CreateLLaVALayers(int imageSize, int channels, int patchSize, int visionHiddenDim, int lmHiddenDim, int numVisionLayers, int numLmLayers, int numHeads, int vocabularySize, int maxSequenceLength) Creates layers for the LLaVA (Large Language and Vision Assistant) native mode.
IEnumerable<ILayer<T>> CreateMMDiTNoisePredictorLayers(int inputChannels, int hiddenSize, int numJointLayers, int numSingleLayers, int numHeads, int patchSize, int contextDim, double mlpRatio) Creates layers for the MMDiT (Multi-Modal Diffusion Transformer) noise predictor.
IEnumerable<ILayer<T>> CreateMOMENTClassificationHead(int numPatches, int hiddenDim, int numClasses) Builds MOMENT’s classification head per Goswami et al.
IEnumerable<ILayer<T>> CreateMOMENTReconstructionHead(int numPatches, int hiddenDim, int patchLength) Builds MOMENT’s per-patch reconstruction head per Goswami et al.
IEnumerable<ILayer<T>> CreateMamba2Layers(int vocabSize, int modelDimension, int numLayers, int stateDimension, int numHeads, int maxSeqLength) Creates layers for the Mamba-2 language model architecture.
IEnumerable<ILayer<T>> CreateMambaLayers(int vocabSize, int modelDimension, int numLayers, int stateDimension, int expandFactor, int maxSeqLength) Creates layers for the Mamba language model architecture.
IEnumerable<ILayer<T>> CreateMask2FormerDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for Mask2Former’s transformer decoder with masked cross-attention.
IEnumerable<ILayer<T>> CreateMask2FormerEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for Mask2Former’s backbone (Swin/ResNet + pixel decoder).
IEnumerable<ILayer<T>> CreateMaskAdapterDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the MaskAdapter model.
IEnumerable<ILayer<T>> CreateMaskAdapterEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the MaskAdapter model.
IEnumerable<ILayer<T>> CreateMaskDINODecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the mask decoder layers for Mask DINO.
IEnumerable<ILayer<T>> CreateMaskDINOEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates the backbone encoder layers for Mask DINO.
IEnumerable<ILayer<T>> CreateMedNeXtDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the MedNeXt model.
IEnumerable<ILayer<T>> CreateMedNeXtEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the MedNeXt model.
IEnumerable<ILayer<T>> CreateMedSAMDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the MedSAM model.
IEnumerable<ILayer<T>> CreateMedSAMEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the MedSAM model.
IEnumerable<ILayer<T>> CreateMedSegDiffV2DecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the MedSegDiffV2 model.
IEnumerable<ILayer<T>> CreateMedSegDiffV2EncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the MedSegDiffV2 model.
IEnumerable<ILayer<T>> CreateMedSegDiffV2SegmentationDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the MedSegDiffV2Segmentation model.
IEnumerable<ILayer<T>> CreateMedSegDiffV2SegmentationEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the MedSegDiffV2Segmentation model.
IEnumerable<ILayer<T>> CreateMixedQueryTransformerDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the query-meld decoder layers for MixedQueryTransformer.
IEnumerable<ILayer<T>> CreateMixedQueryTransformerEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates the backbone encoder layers for MixedQueryTransformer.
IEnumerable<ILayer<T>> CreateMobileSAMDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the MobileSAM model.
IEnumerable<ILayer<T>> CreateMobileSAMEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the MobileSAM model.
IEnumerable<ILayer<T>> CreateNeRFLayers(int positionEncodingLevels, int directionEncodingLevels, int hiddenDim, int numLayers, int colorHiddenDim, int colorNumLayers) Creates layers for the NeRF (Neural Radiance Field) model.
IEnumerable<ILayer<T>> CreateNeuralNoiseReducerLayers(int fftSize, int baseFilters, int numStages, int bottleneckDim, int convKernelSize, int convStride, int convPadding) Creates NeuralNoiseReducer encoder, bottleneck, and decoder layers.
IEnumerable<ILayer<T>> CreateNnUNetDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the NnUNet model.
IEnumerable<ILayer<T>> CreateNnUNetEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the NnUNet model.
IEnumerable<ILayer<T>> CreateODISEDecoderLayers(int[] channelDims, int decoderDim, int numClasses) Builds the ODISE decoder: an SD-U-Net-style transposed-conv upsampling ladder that reverses the encoder’s stride schedule to restore the spatial bottleneck back to full input resolution for dense per-pixel prediction (Xu et al.
IEnumerable<ILayer<T>> CreateODISEEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the ODISE model.
IEnumerable<ILayer<T>> CreateODISESegmentationDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the ODISESegmentation model.
IEnumerable<ILayer<T>> CreateODISESegmentationEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the ODISESegmentation model.
IEnumerable<ILayer<T>> CreateOMGLLaVADecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the OMGLLaVA model.
IEnumerable<ILayer<T>> CreateOMGLLaVAEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the OMGLLaVA model.
IEnumerable<ILayer<T>> CreateOMGSegDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the task-query decoder layers for OMG-Seg.
IEnumerable<ILayer<T>> CreateOMGSegEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates the shared backbone encoder layers for OMG-Seg.
IEnumerable<ILayer<T>> CreateOneFormerDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for OneFormer.
IEnumerable<ILayer<T>> CreateOneFormerEncoderLayers(int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropPathRate, int[] numHeads, int windowSize, int patchSize, int mlpRatio) Creates encoder layers for OneFormer’s backbone with text-conditioned features.
IEnumerable<ILayer<T>> CreateOpenVocabSAMDecoderLayers(int encoderOutputChannels, int neckEmbeddingDim, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the OpenVocabSAM model.
IEnumerable<ILayer<T>> CreateOpenVocabSAMEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the OpenVocabSAM model.
IEnumerable<ILayer<T>> CreatePIDNetDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the PIDNet model.
IEnumerable<ILayer<T>> CreatePIDNetEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the PIDNet model.
IEnumerable<ILayer<T>> CreatePixelLMDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the PixelLM model.
IEnumerable<ILayer<T>> CreatePixelLMEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the PixelLM model.
IEnumerable<ILayer<T>> CreateProPainterLayers(int channels, int height, int width, int numFeatures, int numTransformerBlocks, int numHeads) Creates layers for the ProPainter video inpainting model.
IEnumerable<ILayer<T>> CreateQueryMeldNetDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the query-meld decoder layers for QueryMeldNet.
IEnumerable<ILayer<T>> CreateQueryMeldNetEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates the backbone encoder layers for QueryMeldNet.
IEnumerable<ILayer<T>> CreateRAFTLayers(int channels, int height, int width, int numFeatures, int correlationLevels, int correlationRadius) Creates layers for the RAFT optical flow model.
IEnumerable<ILayer<T>> CreateRIFELayers(int channels, int height, int width, int numFeatures, int numFlowBlocks) Creates layers for the RIFE frame interpolation model.
IEnumerable<ILayer<T>> CreateRWKV4Layers(int vocabSize, int modelDimension, int numLayers, int maxSeqLength) Creates layers for the RWKV-4 language model architecture.
IEnumerable<ILayer<T>> CreateRWKV7Layers(int vocabSize, int modelDimension, int numLayers, int numHeads, double ffnMultiplier, int maxSeqLength) Creates layers for the RWKV-7 language model architecture.
IEnumerable<ILayer<T>> CreateRecurrentGemmaLayers(int vocabSize, int modelDimension, int numLayers, int maxSeqLength) Creates layers for the RecurrentGemma language model architecture.
IEnumerable<ILayer<T>> CreateRepViTSAMDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the RepViTSAM model.
IEnumerable<ILayer<T>> CreateRepViTSAMEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the RepViTSAM model.
IEnumerable<ILayer<T>> CreateResNetBottleneckEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths) Paper-faithful ResNet backbone (He et al.
IEnumerable<ILayer<T>> CreateSAM2ImageEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures) Creates the image encoder layers for SAM2 (Segment Anything Model 2).
IEnumerable<ILayer<T>> CreateSAM2IoUHead(int numFeatures, int featureHeight, int featureWidth, int numMaskCandidates) Creates the IoU (Intersection over Union) prediction head for SAM2.
IEnumerable<ILayer<T>> CreateSAM2MaskDecoderLayers(int numFeatures, int featureHeight, int featureWidth) Creates the shared mask decoder refinement layers for SAM2.
IEnumerable<ILayer<T>> CreateSAM2MaskHead(int numFeatures, int featureHeight, int featureWidth, int numMaskCandidates) Creates the mask prediction head for SAM2.
IEnumerable<ILayer<T>> CreateSAM2MemoryLayers(int numFeatures, int featureHeight, int featureWidth) Creates the memory attention layers for SAM2 temporal consistency.
IEnumerable<ILayer<T>> CreateSAM2OcclusionHead(int numFeatures, int featureHeight, int featureWidth) Creates the occlusion prediction head for SAM2.
IEnumerable<ILayer<T>> CreateSAM2PromptEncoderLayers(int numFeatures, int maskHeight, int maskWidth) Creates the prompt encoder layers for SAM2 (point, box, and mask prompts).
IEnumerable<ILayer<T>> CreateSAMHQDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the HQ mask decoder layers for SAM-HQ.
IEnumerable<ILayer<T>> CreateSAMHQEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates the ViT encoder layers for SAM-HQ (High-Quality Segment Anything).
IEnumerable<ILayer<T>> CreateSANDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the SAN model.
IEnumerable<ILayer<T>> CreateSANEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the SAN model.
IEnumerable<ILayer<T>> CreateSEDDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the SED model.
IEnumerable<ILayer<T>> CreateSEDEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the SED model.
IEnumerable<ILayer<T>> CreateSEEMDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the SEEM model.
IEnumerable<ILayer<T>> CreateSEEMEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the SEEM model.
IEnumerable<ILayer<T>> CreateSambaLayers(int vocabSize, int modelDimension, int numLayers, int stateDimension, int attentionInterval, int maxSeqLength) Creates layers for the Samba hybrid language model architecture.
IEnumerable<ILayer<T>> CreateSegFormerDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the MLP decode head layers for SegFormer.
IEnumerable<ILayer<T>> CreateSegFormerEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] embedDims, int[] depths, int[] numHeads, double dropRate) Creates the Mix Transformer (MiT) encoder layers for SegFormer.
IEnumerable<ILayer<T>> CreateSegGPTDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the SegGPT model.
IEnumerable<ILayer<T>> CreateSegGPTEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the SegGPT model.
IEnumerable<ILayer<T>> CreateSegMambaDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the SegMamba model.
IEnumerable<ILayer<T>> CreateSegMambaEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the SegMamba model.
IEnumerable<ILayer<T>> CreateSileroVadLayers(int frameSize, int convFilters, int numLstmLayers, int lstmHiddenDim, int conv1KernelSize, int conv1Stride, int conv1Padding, int conv2KernelSize, int conv2Stride, int conv2Padding) Creates SileroVad convolutional, LSTM, and output layers.
IEnumerable<ILayer<T>> CreateSimpleVideoSuperResolutionLayers(int inputChannels, int inputHeight, int inputWidth, int scaleFactor) Creates a simple super-resolution architecture for testing and lightweight use.
IEnumerable<ILayer<T>> CreateSlimSAMDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the SlimSAM model.
IEnumerable<ILayer<T>> CreateSlimSAMEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the SlimSAM model.
IEnumerable<ILayer<T>> CreateSlowFastFastPathwayLayers(int inputChannels, int inputHeight, int inputWidth, int fastChannels) Creates the fast pathway layers for SlowFast video recognition.
IEnumerable<ILayer<T>> CreateSlowFastFusionLayers(int slowChannels, int fastChannels, int featureHeight, int featureWidth, int numClasses) Creates the fusion and classification layers for SlowFast.
IEnumerable<ILayer<T>> CreateSlowFastSlowPathwayLayers(int inputChannels, int inputHeight, int inputWidth, int slowChannels) Creates the slow pathway layers for SlowFast video recognition.
IEnumerable<ILayer<T>> CreateSpeechEmotionRecognizerLayers(int numMels, int numFrames, int baseFilters, int numConvBlocks, int hiddenDim, double dropoutRate, int numEmotions, int convKernelSize, int convStride, int convPadding, int poolSize, int poolStride) Creates SpeechEmotionRecognizer convolutional, dense, and output layers.
IEnumerable<ILayer<T>> CreateStableVideoDiffusionLayers(int channels, int height, int width, int latentDim, int numFrames, int vaeChannels, int textEncoderDim, int textEncoderLayers) Creates layers for the Stable Video Diffusion model.
IEnumerable<ILayer<T>> CreateStandardVAEDecoderLayers(int inputChannels, int latentChannels, int baseChannels, int[] channelMultipliers, int numResBlocksPerLevel, int inputHeight, int inputWidth) Creates decoder layers for the StandardVAE (Stable Diffusion VAE architecture).
IEnumerable<ILayer<T>> CreateStandardVAEEncoderLayers(int inputChannels, int latentChannels, int baseChannels, int[] channelMultipliers, int numResBlocksPerLevel, int inputHeight, int inputWidth) Creates encoder layers for the StandardVAE (Stable Diffusion VAE architecture).
IEnumerable<ILayer<T>> CreateSwinUNETRDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the SwinUNETR model.
IEnumerable<ILayer<T>> CreateSwinUNETREncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the SwinUNETR model.
IEnumerable<ILayer<T>> CreateSynthesisVideoStabilizationLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures) Creates a length-preserving conv encoder-decoder for SYNTHESIS-based video stabilization (full-frame neural rendering / multi-frame fusion — e.g.
IEnumerable<ILayer<T>> CreateTemporalVAEDecoderLayers(int inputChannels, int latentChannels, int baseChannels, int[] channelMultipliers, int numResBlocksPerLevel, int inputHeight, int inputWidth) Creates decoder layers for the TemporalVAE (Stable Video Diffusion VAE architecture).
IEnumerable<ILayer<T>> CreateTemporalVAEEncoderLayers(int inputChannels, int latentChannels, int baseChannels, int[] channelMultipliers, int numResBlocksPerLevel, int inputHeight, int inputWidth) Creates encoder layers for the TemporalVAE (Stable Video Diffusion VAE architecture).
IEnumerable<ILayer<T>> CreateTransUNetDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the TransUNet model.
IEnumerable<ILayer<T>> CreateTransUNetEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the TransUNet model.
IEnumerable<ILayer<T>> CreateTransformerEmbeddingLayers(int vocabSize, int embeddingDimension, int maxSequenceLength, int numLayers, int numHeads, int feedForwardDim) Creates layers for the TransformerEmbeddingNetwork.
IEnumerable<ILayer<T>> CreateU2SegDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the mask decoder layers for U2Seg.
IEnumerable<ILayer<T>> CreateU2SegEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates the Swin-T encoder layers for U2Seg (unsupervised segmentation).
IEnumerable<ILayer<T>> CreateUMambaDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the UMamba model.
IEnumerable<ILayer<T>> CreateUMambaEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the UMamba model.
IEnumerable<ILayer<T>> CreateUNINEXTDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the query-based decoder layers for UNINEXT.
IEnumerable<ILayer<T>> CreateUNINEXTEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates the backbone encoder layers for UNINEXT.
IEnumerable<ILayer<T>> CreateUNetNoisePredictorDecoderLayers(int outputChannels, int baseChannels, int[] channelMultipliers, int numResBlocks, int inputHeight, int inputWidth) Creates decoder layers for the UNet noise predictor (Stable Diffusion architecture).
IEnumerable<ILayer<T>> CreateUNetNoisePredictorEncoderLayers(int inputChannels, int baseChannels, int[] channelMultipliers, int numResBlocks, int contextDim, int numHeads, int inputHeight, int inputWidth) Creates encoder layers for the UNet noise predictor (Stable Diffusion architecture).
IEnumerable<ILayer<T>> CreateUViTNoisePredictorLayers(int inputChannels, int hiddenSize, int numLayers, int numHeads, int patchSize) Creates layers for the UViT (U-ViT) noise predictor with skip connections.
IEnumerable<ILayer<T>> CreateUniVSDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the UniVS model.
IEnumerable<ILayer<T>> CreateUniVSEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the UniVS model (Li et al.
IEnumerable<ILayer<T>> CreateUniverSegDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the UniverSeg model.
IEnumerable<ILayer<T>> CreateUniverSegEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the UniverSeg model.
IEnumerable<ILayer<T>> CreateVITSLayers(int hiddenDim, int numEncoderLayers, int numHeads, int maxPhonemeLength, int numFlowLayers, int phonemeVocabSize, int[] upsampleRates) Creates VITS text encoder, duration predictor, flow, and decoder layers.
IEnumerable<ILayer<T>> CreateVMambaDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the VMamba model.
IEnumerable<ILayer<T>> CreateVMambaEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the VMamba model.
IEnumerable<ILayer<T>> CreateViMUNetDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the ViMUNet model.
IEnumerable<ILayer<T>> CreateViMUNetEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the ViMUNet model.
IEnumerable<ILayer<T>> CreateViTAdapterDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the decoder layers for ViT-Adapter.
IEnumerable<ILayer<T>> CreateViTAdapterEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int embedDim, int[] depths, int[] numHeads, double dropRate) Creates the ViT-Adapter encoder layers.
IEnumerable<ILayer<T>> CreateViTCoMerDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the decoder layers for ViT-CoMer.
IEnumerable<ILayer<T>> CreateViTCoMerEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int embedDim, int[] cnnChannels, int[] depths, double dropRate) Creates the ViT-CoMer hybrid encoder layers combining CNN and transformer features.
IEnumerable<ILayer<T>> CreateViTCoMerLayers(int inputChannels, int inputHeight, int inputWidth, int embedDim, int[] cnnChannels, int[] depths, int decoderDim, int numClasses, double dropRate) Creates the complete native ViT-CoMer graph with parallel ViT/CNN branches, MRFP refinement, bidirectional CTI, and the multi-scale segmentation decoder.
IEnumerable<ILayer<T>> CreateVideoCLIPLayers(int channels, int height, int width, int hiddenDim, int embeddingDim, int numSpatialBlocks, int numTemporalBlocks, int numTextBlocks, int numFrames, int textMaxLength) Creates layers for the VideoCLIP video-text understanding model.
IEnumerable<ILayer<T>> CreateVideoCLIPLayers(int imageSize, int channels, int patchSize, int visionHiddenDim, int textHiddenDim, int embeddingDimension, int numFrameEncoderLayers, int numTemporalLayers, int numTextLayers, int numHeads, int vocabularySize, int maxSequenceLength) Creates layers for the VideoCLIP native mode.
IEnumerable<ILayer<T>> CreateVideoGigaGANLayers(int inputChannels, int inputHeight, int inputWidth, int numFeatures, int numResBlocks, int numStyleLayers, int scaleFactor, int flowPyramidLevels, double hfShuttleWeight) Creates the native VideoGigaGAN generator described by Xu et al.
IEnumerable<ILayer<T>> CreateVideoLISADecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the VideoLISA model.
IEnumerable<ILayer<T>> CreateVideoLISAEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the VideoLISA model.
IEnumerable<ILayer<T>> CreateVideoUNetPredictorDecoderLayers(int outputChannels, int baseChannels, int[] channelMultipliers, int numResBlocks, int inputHeight, int inputWidth) Creates decoder layers for the VideoUNet noise predictor.
IEnumerable<ILayer<T>> CreateVideoUNetPredictorEncoderLayers(int inputChannels, int baseChannels, int[] channelMultipliers, int numResBlocks, int contextDim, int numHeads, int numTemporalLayers, int inputHeight, int inputWidth) Creates layers for the VideoUNet noise predictor (Stable Video Diffusion architecture).
IEnumerable<ILayer<T>> CreateVisionTransformerLayers(int imageHeight, int imageWidth, int channels, int patchSize, int hiddenDim, int numLayers, int numHeads, int mlpDim, int numClasses) Creates layers for the Vision Transformer (ViT) architecture.
IEnumerable<ILayer<T>> CreateXDecoderDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates the dual-path decoder layers for X-Decoder (pixel + token paths).
IEnumerable<ILayer<T>> CreateXDecoderEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates the Focal transformer encoder layers for X-Decoder.
IEnumerable<ILayer<T>> CreateXLSTMLayers(int vocabSize, int modelDimension, int numLayers, int numHeads, int maxSeqLength) Creates layers for the xLSTM language model architecture.
IEnumerable<ILayer<T>> CreateYOLO11SegDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the YOLO11Seg model.
IEnumerable<ILayer<T>> CreateYOLO11SegEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the YOLO11Seg model.
IEnumerable<ILayer<T>> CreateYOLO26SegDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the YOLO26Seg model.
IEnumerable<ILayer<T>> CreateYOLO26SegEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the YOLO26Seg model.
IEnumerable<ILayer<T>> CreateYOLOv12SegDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the YOLOv12Seg model.
IEnumerable<ILayer<T>> CreateYOLOv12SegEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the YOLOv12Seg model.
IEnumerable<ILayer<T>> CreateYOLOv8SegDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the YOLOv8-Seg model.
IEnumerable<ILayer<T>> CreateYOLOv8SegEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the YOLOv8-Seg model.
IEnumerable<ILayer<T>> CreateYOLOv9SegDecoderLayers(int encoderOutputChannels, int decoderDim, int numClasses, int featureHeight, int featureWidth) Creates decoder layers for the YOLOv9Seg model.
IEnumerable<ILayer<T>> CreateYOLOv9SegEncoderLayers(int inputChannels, int inputHeight, int inputWidth, int[] channelDims, int[] depths, double dropRate) Creates encoder layers for the YOLOv9Seg model.
IEnumerable<ILayer<T>> CreateZamba2Layers(int vocabSize, int modelDimension, int numLayers, int stateDimension, int numHeads, int attentionInterval, int maxSeqLength) Creates layers for the Zamba2 hybrid language model architecture.
IEnumerable<ILayer<T>> CreateZambaLayers(int vocabSize, int modelDimension, int numLayers, int stateDimension, int attentionInterval, int maxSeqLength) Creates layers for the Zamba hybrid language model architecture.
ValueTuple<IEnumerable<ILayer<T>>, IEnumerable<ILayer<T>>> CreateDefaultDeepOperatorNetworkLayers(int branchInputSize, int trunkInputSize, int outputSize, int hiddenLayerCount, int hiddenLayerSize) Creates default layers for a Deep Operator Network (DeepONet).
ValueTuple<IEnumerable<ILayer<T>>, IEnumerable<ILayer<T>>> CreateDefaultDessurtLayers(int encoderDim, int decoderDim, int encoderLayers, int decoderLayers, int numHeads, int vocabSize) Creates default Dessurt (self-supervised document transformer) layers.
ValueTuple<IEnumerable<ILayer<T>>, IEnumerable<ILayer<T>>> CreateDefaultDonutLayers(int imageHeight, int imageWidth, int inputChannels, int embedDim, int[] depths, int[] numHeads, int windowSize, int patchSize, int mlpRatio, int decoderHiddenDim, int numDecoderLayers, int decoderHeads, int vocabSize, int maxGenerationLength) Creates default Donut layers for OCR-free document understanding.
ValueTuple<IEnumerable<ILayer<T>>, IEnumerable<ILayer<T>>> CreateDefaultMATCHALayers(int encoderDim, int decoderDim, int encoderLayers, int decoderLayers, int numHeads, int vocabSize, int maxPatchesPerImage) Creates default MATCHA (chart understanding) layers.
int ComputeProprietaryAPIEncoderBoundary(int numVisionLayers, double dropoutRate) Computes the encoder/decoder split index for a layer stack built by CreateDefaultProprietaryAPILayers — the count of encoder-side layers, i.e.

Fields

Field Summary
int AlphaFactorExtractorLayerCount Layers in the AlphaFactorModel factor extractor span: the feature encoder, the hidden block, and the factor head with its normalization.
int AlphaFactorPredictorLayerCount Layers in the AlphaFactorModel alpha predictor span.
int FactorVAEAlphaLayerCount Layers in the FactorVAE alpha (idiosyncratic return) head.
int FactorVAEBetaLayerCount Layers in the FactorVAE beta (factor exposure) head.
int FactorVAEDecoderLayerCount Number of layers in the FactorVAE decoder span.
int FactorVAEFeatureExtractorLayerCount Number of layers in the FactorVAE feature extractor span.
int FactorVAEPosteriorLayerCount Number of layers in the FactorVAE posterior (factor encoder) span.
int FactorVAEPriorLayerCount Number of layers in the FactorVAE prior (factor predictor) span.
int RoomImpulseResponseFixedLayers Layers emitted by CreateDefaultRoomImpulseResponseLayers outside the encoder and decoder blocks: the 3-layer MLP plus the three output heads.
int RoomImpulseResponseLayersPerDecoderBlock Layers emitted per decoder block by CreateDefaultRoomImpulseResponseLayers.
int RoomImpulseResponseLayersPerEncoderBlock Layers emitted per encoder block by CreateDefaultRoomImpulseResponseLayers.