Here is a list of all functions with links to the classes they belong to:

- s -

• sampleFromLogits() : Mila::Dnn::GptModel< TDeviceType, TPrecision >, Mila::Dnn::LlamaModel< TDeviceType, TPrecision >
• sampleToken() : Mila::Dnn::GptModel< TDeviceType, TPrecision >, Mila::Dnn::LlamaModel< TDeviceType, TPrecision >
• save() : Mila::Data::BpeVocabulary, Mila::Data::CharVocabulary, Mila::Data::TokenizerVocabulary, Mila::Dnn::Dropout< TDeviceType, TInput, TOutput >, Mila::Dnn::Network< TDeviceType, TPrecision >
• save_() : Mila::Dnn::Component< TDeviceType, TPrecision >, Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >, Mila::Dnn::FusedComponent< TDeviceType, TPrecision >, Mila::Dnn::Gelu< TDeviceType, TPrecision >, Mila::Dnn::GptBlock< TDeviceType, TPrecision >, Mila::Dnn::GptTransformer< TDeviceType, TPrecision >, Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >, Mila::Dnn::LayerNorm< TDeviceType, TPrecision >, Mila::Dnn::Linear< TDeviceType, TComputePrecision, TWeightQuant >, Mila::Dnn::LlamaBlock< TDeviceType, TPrecision, TWeightQuant, TKvPolicy >, Mila::Dnn::LlamaTransformer< TDeviceType, TPrecision, TWeightQuantization, TKvCachePolicy >, Mila::Dnn::Lpe< TDeviceType, TIndex, TPrecision >, Mila::Dnn::MLP< TDeviceType, TPrecision >, Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >, Mila::Dnn::Network< TDeviceType, TPrecision >, Mila::Dnn::Residual< TDeviceType, TPrecision >, Mila::Dnn::RmsNorm< TDeviceType, TPrecision >, Mila::Dnn::Rope< TDeviceType, TPrecision >, Mila::Dnn::Softmax< TDeviceType, TPrecision >, Mila::Dnn::SoftmaxCrossEntropy< TDeviceType, TLogits, TTargets, TPrecision >, Mila::Dnn::Swiglu< TDeviceType, TPrecision >, Mila::Dnn::TokenEmbedding< TDeviceType, TIndex, TPrecision >
• saveComponentGraph() : Mila::Dnn::Network< TDeviceType, TPrecision >
• saveContent() : Mila::Data::BpeVocabulary, Mila::Data::CharVocabulary
• saveNetworkMetadata() : Mila::Dnn::Network< TDeviceType, TPrecision >
• scalesDuringInference() : Mila::Dnn::DropoutConfig
• scopedPath() : Mila::Dnn::Serialization::ModelArchive
• ScopedScope() : Mila::Dnn::Serialization::ModelArchive::ScopedScope
• sequenceLength() : Mila::Data::TokenSequenceLoader< TMemoryResource >
• SerializationMetadata() : Mila::Data::SerializationMetadata, Mila::Dnn::Optimizers::SerializationMetadata, Mila::Dnn::Serialization::SerializationMetadata, Mila::Dnn::SerializationMetadata
• set() : Mila::Data::SerializationMetadata, Mila::Dnn::Optimizers::SerializationMetadata, Mila::Dnn::Serialization::SerializationMetadata, Mila::Dnn::SerializationMetadata
• setCurrentBatch() : Mila::Data::DataLoader< TInputDataType, TTargetDataType, TMemoryResource >
• setDefaultLogger() : Mila::Logging::Logger
• setEval() : Mila::Dnn::Model< TDeviceType, TPrecision >
• setExecutionContext() : Mila::Dnn::Component< TDeviceType, TPrecision >
• setGradients() : Mila::Dnn::Compute::CpuAttentionOp, Mila::Dnn::Compute::CpuEncoderOp, Mila::Dnn::Compute::CpuLayerNormOp, Mila::Dnn::Compute::CpuLinearOp, Mila::Dnn::Compute::Cuda::Gqa::CudaGqaOp< TPrecision >, Mila::Dnn::Compute::Cuda::LayerNorm::CudaLayerNormOp< TPrecision >, Mila::Dnn::Compute::Cuda::Linear::CudaLinearOp< TComputePrecision, TWeightQuant >, Mila::Dnn::Compute::Cuda::Lpe::CudaLpeOp< TInput, TPrecision >, Mila::Dnn::Compute::Cuda::MultiHeadAttention::CudaMultiHeadAttentionOp< TPrecision >, Mila::Dnn::Compute::Cuda::RmsNorm::CudaRmsNormOp< TPrecision >, Mila::Dnn::Compute::Cuda::TokenEmbedding::CudaTokenEmbeddingOp< TInput, TPrecision >, Mila::Dnn::Compute::Operation< TDeviceType, TComputePrecision >
• setIncludeSourceLocation() : Mila::Logging::ConsoleSink, Mila::Logging::FileSink
• setIncludeTimestamp() : Mila::Logging::ConsoleSink, Mila::Logging::FileSink
• setLearningRate() : Mila::Dnn::Compute::CpuAdamWOptimizer< TPrecision >, Mila::Dnn::Compute::CudaAdamWOptimizer< TPrecision >, Mila::Dnn::Compute::Optimizer< TDeviceType, TPrecision >, Mila::Dnn::Optimizers::AdamWOptimizer< TDeviceType, TPrecision >
• setLevel() : Mila::Logging::ConsoleSink, Mila::Logging::FileSink, Mila::Logging::Logger, Mila::Logging::NullSink
• setName() : Mila::Dnn::Tensor< TDataType, TMemoryResource >
• setParameters() : Mila::Dnn::Compute::CpuAttentionOp, Mila::Dnn::Compute::CpuEncoderOp, Mila::Dnn::Compute::CpuLayerNormOp, Mila::Dnn::Compute::CpuLinearOp, Mila::Dnn::Compute::CpuSoftmaxCrossEntropyOp< TPrecision, TLogits, TTargets >, Mila::Dnn::Compute::CpuSoftmaxOp, Mila::Dnn::Compute::Cuda::Gqa::CudaGqaOp< TPrecision >, Mila::Dnn::Compute::Cuda::LayerNorm::CudaLayerNormOp< TPrecision >, Mila::Dnn::Compute::Cuda::Linear::CudaLinearOp< TComputePrecision, TWeightQuant >, Mila::Dnn::Compute::Cuda::Lpe::CudaLpeOp< TInput, TPrecision >, Mila::Dnn::Compute::Cuda::MultiHeadAttention::CudaMultiHeadAttentionOp< TPrecision >, Mila::Dnn::Compute::Cuda::RmsNorm::CudaRmsNormOp< TPrecision >, Mila::Dnn::Compute::Cuda::Softmax::CudaSoftmaxOp< TPrecision >, Mila::Dnn::Compute::Cuda::SoftmaxCrossEntropy::CudaSoftmaxCrossEntropyOp< TPrecision, TLogits, TTargets >, Mila::Dnn::Compute::Cuda::TokenEmbedding::CudaTokenEmbeddingOp< TInput, TPrecision >, Mila::Dnn::Compute::Operation< TDeviceType, TComputePrecision >
• setSeed() : Mila::Core::RandomGenerator, Mila::Dnn::Dropout< TDeviceType, TInput, TOutput >
• setState() : Mila::Dnn::Compute::Cuda::Gqa::CudaGqaOp< TPrecision >, Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >, Mila::Dnn::LlamaBlock< TDeviceType, TPrecision, TWeightQuant, TKvPolicy >
• setTrainingMode() : Mila::Dnn::Component< TDeviceType, TPrecision >, Mila::Dnn::Compute::Operation< TDeviceType, TComputePrecision >
• setUId() : Mila::Dnn::Tensor< TDataType, TMemoryResource >
• setWeightDecay() : Mila::Dnn::Compute::CudaAdamWOptimizer< TPrecision >, Mila::Dnn::Optimizers::AdamWOptimizer< TDeviceType, TPrecision >
• setWeightScales() : Mila::Dnn::Compute::Cuda::Linear::CudaLinearOp< TComputePrecision, TWeightQuant >
• setWeightZeroPoints() : Mila::Dnn::Compute::Cuda::Linear::CudaLinearOp< TComputePrecision, TWeightQuant >
• shape() : Mila::Dnn::ITensor, Mila::Dnn::Tensor< TDataType, TMemoryResource >
• shapeToString() : Mila::Dnn::Compute::CpuAdamWOptimizer< TPrecision >, Mila::Dnn::Compute::CudaAdamWOptimizer< TPrecision >
• shouldInitializeParameters() : Mila::Dnn::BuildContext
• shouldNormalizeUnicode() : Mila::Data::CharVocabularyConfig
• shuffleSequenceIndices() : Mila::Data::TokenSequenceLoader< TMemoryResource >
• size() : Mila::Data::SerializationMetadata, Mila::Dnn::Compute::Cuda::CublasLtPlanCache< TPlan >, Mila::Dnn::ITensor, Mila::Dnn::Optimizers::SerializationMetadata, Mila::Dnn::Serialization::SerializationMetadata, Mila::Dnn::SerializationMetadata, Mila::Dnn::Tensor< TDataType, TMemoryResource >, Mila::Dnn::TensorBuffer< TDataType, TMemoryResource, TrackMemory >, Mila::Dnn::TensorShape
• sizeBytes() : Mila::Dnn::Serialization::ITensorBlob, Mila::Dnn::Serialization::TensorBlob< MR >
• Softmax() : Mila::Dnn::Softmax< TDeviceType, TPrecision >
• softmax_backward() : Mila::Dnn::Compute::Cuda::Gqa::Detail::cuda_gqa_kernels< float >, Mila::Dnn::Compute::Cuda::Gqa::Detail::cuda_gqa_kernels< nv_bfloat16 >, Mila::Dnn::Compute::Cuda::MultiHeadAttention::Detail::cuda_mha_kernels< float >, Mila::Dnn::Compute::Cuda::MultiHeadAttention::Detail::cuda_mha_kernels< half >
• softmax_decode_forward() : Mila::Dnn::Compute::Cuda::Gqa::Detail::cuda_gqa_kernels< float >, Mila::Dnn::Compute::Cuda::Gqa::Detail::cuda_gqa_kernels< nv_bfloat16 >, Mila::Dnn::Compute::Cuda::MultiHeadAttention::Detail::cuda_mha_kernels< float >, Mila::Dnn::Compute::Cuda::MultiHeadAttention::Detail::cuda_mha_kernels< half >
• softmax_forward() : Mila::Dnn::Compute::Cuda::Gqa::Detail::cuda_gqa_kernels< float >, Mila::Dnn::Compute::Cuda::Gqa::Detail::cuda_gqa_kernels< nv_bfloat16 >, Mila::Dnn::Compute::Cuda::MultiHeadAttention::Detail::cuda_mha_kernels< float >, Mila::Dnn::Compute::Cuda::MultiHeadAttention::Detail::cuda_mha_kernels< half >
• softmax_padded_forward() : Mila::Dnn::Compute::Cuda::Gqa::Detail::cuda_gqa_kernels< float >, Mila::Dnn::Compute::Cuda::Gqa::Detail::cuda_gqa_kernels< nv_bfloat16 >, Mila::Dnn::Compute::Cuda::MultiHeadAttention::Detail::cuda_mha_kernels< float >, Mila::Dnn::Compute::Cuda::MultiHeadAttention::Detail::cuda_mha_kernels< half >
• SoftmaxCrossEntropy() : Mila::Dnn::SoftmaxCrossEntropy< TDeviceType, TLogits, TTargets, TPrecision >
• sortBytes() : Mila::Data::CharVocabulary
• spirv_bit_width() : Mila::Dnn::VulkanTensorTraits
• split() : Mila::Dnn::Compute::Cuda::Detail::cuda_structural_kernels< float >, Mila::Dnn::Compute::Cuda::Detail::cuda_structural_kernels< nv_bfloat16 >, Mila::Dnn::Compute::Cuda::StructuralOps
• standard() : Mila::Data::SpecialTokens
• start() : Mila::Dnn::Compute::CudaTimer
• step() : Mila::Dnn::Compute::CpuAdamWOptimizer< TPrecision >, Mila::Dnn::Compute::CudaAdamWOptimizer< TPrecision >, Mila::Dnn::Compute::Optimizer< TDeviceType, TPrecision >, Mila::Dnn::Optimizers::AdamWOptimizer< TDeviceType, TPrecision >
• StepLogger() : Mila::Utils::StepLogger
• stop() : Mila::Dnn::Compute::CudaTimer
• stopTokens() : Mila::Dnn::LanguageModel< TDeviceType, TPrecision >, Mila::Dnn::LlamaModel< TDeviceType, TPrecision >
• storageBytes() : Mila::Dnn::TensorBuffer< TDataType, TMemoryResource, TrackMemory >
• strides() : Mila::Dnn::Tensor< TDataType, TMemoryResource >
• subtract() : Mila::Dnn::Compute::Cpu::MathOps, Mila::Dnn::Compute::Cuda::MathOps
• subtractImpl() : Mila::Dnn::Compute::Cuda::MathOps
• sum() : Mila::Dnn::Compute::Cpu::MathOps, Mila::Dnn::Compute::Cuda::MathOps
• sumImpl() : Mila::Dnn::Compute::Cuda::MathOps
• supports() : Mila::Dnn::CpuTensorDataTypeTraits, Mila::Dnn::VulkanTensorTraits
• supportsComputeShaders() : Mila::Dnn::Compute::MetalDevice
• supportsCuBLASLt() : Mila::Dnn::Compute::Cuda::Linear::CudaLinearOp< TComputePrecision, TWeightQuant >
• supportsGPUFamily() : Mila::Dnn::Compute::MetalDevice
• supportsKVCache() : Mila::Dnn::GptBlock< TDeviceType, TPrecision >, Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >, Mila::Dnn::LlamaBlock< TDeviceType, TPrecision, TWeightQuant, TKvPolicy >, Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >
• supportsMixedPrecision() : Mila::Data::DataLoader< TInputDataType, TTargetDataType, TMemoryResource >
• swapBuffers() : Mila::Data::TokenSequenceLoader< TMemoryResource >
• Swiglu() : Mila::Dnn::Swiglu< TDeviceType, TPrecision >
• SwigluConfig() : Mila::Dnn::SwigluConfig
• synchronize() : Mila::Dnn::Component< TDeviceType, TPrecision >, Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >, Mila::Dnn::Compute::ExecutionContext< DeviceType::Cpu >, Mila::Dnn::Compute::ExecutionContext< DeviceType::Cuda >, Mila::Dnn::Compute::ExecutionContext< DeviceType::Metal >, Mila::Dnn::Compute::ExecutionContext< DeviceType::Vulkan >, Mila::Dnn::Compute::IExecutionContext, Mila::Dnn::FusedComponent< TDeviceType, TPrecision >, Mila::Dnn::Gelu< TDeviceType, TPrecision >, Mila::Dnn::GroupedQueryAttention< TDeviceType, TComputePrecision, TKvPolicy >, Mila::Dnn::LayerNorm< TDeviceType, TPrecision >, Mila::Dnn::Linear< TDeviceType, TComputePrecision, TWeightQuant >, Mila::Dnn::Lpe< TDeviceType, TIndex, TPrecision >, Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >, Mila::Dnn::Network< TDeviceType, TPrecision >, Mila::Dnn::Residual< TDeviceType, TPrecision >, Mila::Dnn::RmsNorm< TDeviceType, TPrecision >, Mila::Dnn::Rope< TDeviceType, TPrecision >, Mila::Dnn::Softmax< TDeviceType, TPrecision >, Mila::Dnn::SoftmaxCrossEntropy< TDeviceType, TLogits, TTargets, TPrecision >, Mila::Dnn::Swiglu< TDeviceType, TPrecision >, Mila::Dnn::TokenEmbedding< TDeviceType, TIndex, TPrecision >