Class List

Here are the classes, structs, unions and interfaces with brief descriptions:

NMila	Mila main API namespace
NCore
CRandomGenerator	Singleton class providing centralized random number generation
NData
CBpeTokenizer	Unified BPE tokenizer targeting GPT-2, Llama 3.x, and Mistral model families
CBpeTrainer	Corpus accumulator and trainer for BPE vocabularies
CBpeVocabulary	Unified Byte Pair Encoding (BPE) vocabulary
CBpeVocabularyConfig	Configuration for the BPE vocabulary
CCharTokenizer	Character-level tokenizer
CCharTrainer	Character-level tokenizer trainer
CCharVocabulary	Character vocabulary for tokenization
CCharVocabularyConfig	Configuration for Character-level tokenizer training
CDataLoader	Device-agnostic data loader interface using abstract tensor data types
CMilaFileHeader	Common file header for Mila data files
CSerializationMetadata	Type-safe metadata container for component serialization
CSpecialTokens	Configuration for special tokens across all tokenizer types
CTokenizer
CTokenizerTrainer	Abstract interface for training tokenizer vocabularies from text corpora
CTokenizerVocabulary	Generic tokenizer vocabulary interface
CTokenSequenceLoader	Token sequence loader for autoregressive language models
CTokenSequenceLoaderConfig	Configuration for StreamingSequenceLoader behavior
CTrainerFactory	Factory for creating tokenizer trainers and loading vocabularies
NDistribution
CCurlHttpTransport	Libcurl as an IHttpTransport
CFetchReport
CHttpClient	GET a URL, streaming the body to a sink, over whichever transport it is given
CHttpFetch
CHttpHeader
CHttpRequest
CHttpResponse
CHttpResult
CHubModel	One repository as a hub reports it, before any manifest is fetched
CHuggingFaceHub	The HuggingFace hub
CIHttpTransport	Performs one HTTP GET
CIModelHub	A remote that serves manifests and files
CInstallOptions	How a package is to be installed
CModelCoordinate	Where a repository lives on a hub
CModelFile	One file composing a model, as the manifest declares it
CModelManifest	A model's published description
CModelPackage	Everything a published or installable model is, as a directory
CModelRecord	An installed model: what the manifest published, plus how this copy came to be here
CModelResolver	Pulls a published model into the store
CModelStore	The local store of installed models
CNullHttpTransport	The transport of a build compiled without one
CPackageRequest	One model's files, as a caller hands them to the packager
CPackageValidation	What validation found
CPruneOptions
CRemovalReport
CSha256	Incremental SHA-256
CStoredModel	A record together with where its bytes actually are
CStoreUsage
NDnn
NCompute
NCuda
CCudaDataTypeTraits	Compile-time mapping from TensorDataType -> cudaDataType_t
CCudaDataTypeTraits< TensorDataType::BF16 >
CCudaDataTypeTraits< TensorDataType::FP16 >
CCudaDataTypeTraits< TensorDataType::FP32 >
CCudaDataTypeTraits< TensorDataType::FP8_E4M3 >
CCudaDataTypeTraits< TensorDataType::FP8_E5M2 >
CCudaDataTypeTraits< TensorDataType::INT32 >
CCudaDataTypeTraits< TensorDataType::INT8 >
CTensorDataTypeMap	Compile-time mapping from abstract TensorDataType -> CUDA native device type
CTensorDataTypeMap< TensorDataType::BF16 >	Maps TensorDataType::BF16 to CUDA __nv_bfloat16
CTensorDataTypeMap< TensorDataType::FP16 >	Maps TensorDataType::FP16 to CUDA __half
CTensorDataTypeMap< TensorDataType::FP32 >	Maps TensorDataType::FP32 to CUDA float
CTensorDataTypeMap< TensorDataType::FP4_E2M1 >	Maps TensorDataType::FP4_E2M1 to std::uint8_t
CTensorDataTypeMap< TensorDataType::FP4_E3M0 >	Maps TensorDataType::FP4_E3M0 to std::uint8_t
CTensorDataTypeMap< TensorDataType::FP8_E4M3 >	Maps TensorDataType::FP8_E4M3 to CUDA __nv_fp8_e4m3
CTensorDataTypeMap< TensorDataType::FP8_E5M2 >	Maps TensorDataType::FP8_E5M2 to CUDA __nv_fp8_e5m2
CTensorDataTypeMap< TensorDataType::INT16 >	Maps TensorDataType::INT16 to std::int16_t
CTensorDataTypeMap< TensorDataType::INT32 >	Maps TensorDataType::INT32 to std::int32_t
CTensorDataTypeMap< TensorDataType::INT8 >	Maps TensorDataType::INT8 to std::int8_t
CTensorDataTypeMap< TensorDataType::UINT16 >	Maps TensorDataType::UINT16 to std::uint16_t
CTensorDataTypeMap< TensorDataType::UINT32 >	Maps TensorDataType::UINT32 to std::uint32_t
CTensorDataTypeMap< TensorDataType::UINT8 >	Maps TensorDataType::UINT8 to std::uint8_t
CCpuDevice	Class representing a CPU compute device
CCpuDeviceRegistrar	CPU device plugin for device-agnostic registration
CCpuMemoryResource	CPU memory resource for host-accessible memory allocation
CCudaDevice	Class representing a CUDA compute device instance
CCudaDeviceMemoryResource	CUDA device memory resource for GPU-accessible memory allocation
CCudaDeviceProps	Wrapper for CUDA device properties with cached values
CCudaDeviceRegistrar	CUDA device registrar for device-agnostic registration
CCudaManagedMemoryResource	CUDA managed memory resource for unified host/device accessible memory
CCudaPinnedMemoryResource	CUDA pinned memory resource for fast host/device transfer memory
CCudaTimer	GPU-accurate interval timer using a CUDA event pair
CDevice	Abstract interface for compute device implementations
CDeviceAccessible
CDeviceConstructionKey	Construction key for device factories
CDeviceId	Lightweight identifier for a compute device
CDeviceMemoryInfo	What a device currently has free, and what it has in total
CDeviceRegistrar	Device-agnostic registrar for automatic device discovery and registration
CDeviceRegistry	Registry of discovered compute devices with lazy instantiation
CDeviceTypeTraits
CDeviceTypeTraits< DeviceType::Cpu >	DeviceTypeTraits specialization for the CPU device
CDeviceTypeTraits< DeviceType::Cuda >	DeviceTypeTraits specialization for the CUDA device
CExecutionContext	Templated execution context for device-specific operations
CExecutionContext< DeviceType::Cpu >	CPU execution context specialization
CExecutionContext< DeviceType::Cuda >	CUDA execution context specialization
CExecutionContext< DeviceType::Metal >	Metal execution context specialization
CExecutionContext< DeviceType::Vulkan >	Vulkan execution context specialization
CGqaState	Non-owning pointers to shared transient GQA scratch buffers
CHostAccessible
CIExecutionContext	Type-erased execution context interface
CIKvCacheLifecycle	Capability interface for KV-cache state management
CIKvInference	Compute interface for attention operations that maintain a KV cache
CIPackedKvInference	KV-cache inference interface for packed-QKV MHA backends
CIPositionalDecode	Capability interface for position-dependent unary operations
CIPositionalPairedOp	Capability interface for position-dependent paired operations
CMemoryAllocationStats	Process-wide allocator counters for all TrackedMemoryResource instances
CMemoryResource	Clean memory resource abstraction for device-specific memory allocation
CMemoryResourceTraits	Memory resource traits for compile-time dispatch optimization
CMemoryResourceTraits< CpuMemoryResource >	CPU-specific memory resource traits providing detailed CPU backend characteristics
CMemoryResourceTraits< CudaDeviceMemoryResource >	CUDA device memory resource traits providing detailed GPU backend characteristics
CMemoryResourceTraits< CudaManagedMemoryResource >	CUDA managed memory resource traits providing unified memory characteristics
CMemoryResourceTraits< CudaPinnedMemoryResource >	CUDA pinned memory resource traits providing fast transfer characteristics
CMetalDevice	Class representing a Metal compute device instance
CMetalDevicePlugin	Metal device plugin for device-agnostic registration
CMetalMemoryResource	Stub implementation for non-Apple platforms
COperation
COperationTraits	Primary traits template for unified compile-time operation dispatch
CTrackedMemoryResource	A memory resource wrapper that tracks allocation and deallocation statistics
CVulkanDevice	Class representing a Vulkan compute device instance
CVulkanMemoryResource	Stub implementation for platforms without Vulkan support
NGemma
CGemmaToolCall	A tool call parsed out of the model's native <|tool_call> emission
NOptimizers
CAdamWConfig	Configuration for AdamW optimizer
CAdamWOptimizer	Device-agnostic AdamW optimizer
CSerializationMetadata	Type-safe metadata container for component serialization
NQuant
NKvCache
CNoKvCompression	Identity policy - no compression
CPerChannelKvFp8	Symmetric per-head per-token FP8 KV cache compression policy
CSlidingWindowKvCache	Bounded sliding-window KV cache (uncompressed ring buffer)
NWeight
CNoWeightQuant
CPerChannelFp8
CPerGroupFp4
CPerGroupInt4
NSerialization
CArchiveSerializer	Interface for hierarchical archive serializers
CITensorBlob	Type-erased interface for a serialized tensor blob
CModelArchive	ModelArchive provides high-level helpers for component serialization
CScopedScope
CPretrainedMetadata	Metadata for pretrained model
CPretrainedModelReader	Reader for Mila pretrained binary format
CSafeTensorsWriter	Writes a safetensors file: u64 header length, JSON header, packed data
CSerializationMetadata	Type-safe metadata container for component serialization
CSerializer	Minimal base interface for model serialization backends
CTensorBlob	Concrete tensor blob owning a TensorBuffer-backed raw byte buffer
CTensorBlobMetadata	Metadata for a tensor blob in pretrained model format
CTensorBlobView	Non-owning ITensorBlob view over externally-owned bytes
CTensorMetadata	Metadata describing a tensor in serialized form
CZipSerializer	ZIP archive serializer built on miniz
CActivation	Unified elementwise activation component
CActivationConfig	Configuration class for the elementwise Activation component
CAxisPartition	Information about axis partitioning of a tensor
CBufferedTokenStreamer	Buffers BufSize tokens before forwarding a contiguous span to Sink
CBuildContext	Build-time context for Component::build()
CComponent	Abstract base class for neural network components
CComponentConfig	Abstract base for component configuration objects
CComponentFactory	Factory for reconstructing components from serialized archives
CCompositeComponent	A component that contains and manages child components
CConstantLRScheduler	Constant learning-rate scheduler
CCosineLRScheduler	Cosine annealing scheduler
CCpuTensorDataTypeTraits	CPU-specific traits for abstract tensor data types
CCrossEntropyConfig	Configuration for fused SoftmaxCrossEntropy loss
CDecoder
Cdependent_false
Cfunctor_of	Compile-time map from an elementwise ActivationType to its functor
Cfunctor_of< ActivationType::Gelu >
Cfunctor_of< ActivationType::LeakyRelu >
Cfunctor_of< ActivationType::Mish >
Cfunctor_of< ActivationType::None >
Cfunctor_of< ActivationType::Relu >
Cfunctor_of< ActivationType::Sigmoid >
Cfunctor_of< ActivationType::Silu >
Cfunctor_of< ActivationType::Tanh >
CGatedMLP	Gated feed-forward (GatedMLP) composite component
CGatedMLPConfig	Configuration class for the gated feed-forward (GatedMLP) block
CGelu	Gaussian Error Linear Unit (GELU) activation component
CGeluConfig	Configuration class for GELU module
CGemmaBlock	One Gemma 4 decoder block; kGlobal selects the global (full-attention) geometry
CBlockBuildContexts	The per-child build contexts and split-scratch geometry this block implies
CGemmaBlockWorkspace	Transformer-owned shared activation workspace for GemmaBlock (pooling)
CGemmaConfig	Network-level configuration for Gemma 4 transformer networks
CGemmaModel	Gemma 4 compatible inference model
CGemmaModelConfig	Deployment configuration for Gemma language models
CGemmaTransformer	Gemma 4 transformer (decoder-only) for autoregressive inference
CGenerateParams	Per-call inputs to LanguageModel::generate
CGptBlock	Transformer encoder block as a composite component
CGptBlockConfig	Configuration class for GPT transformer blocks
CGptConfig	Network-level configuration for GPT-style transformer networks
CGptModel	GPT inference model
CGptModelConfig	Deployment configuration for Gpt language models
CGptTransformer	GPT-2 style transformer (decoder-only) for autoregressive token prediction
CGqaConfig	Configuration class for the Grouped-Query Attention module
CGroupedQueryAttention	Grouped-Query Attention module that accepts concatenated QKV input
CIDecoderLayer	Polymorphic inference interface for one decoder layer
CITensor	Abstract interface providing essential tensor information and data access
CLanguageModel
CLanguageModelConfig	CRTP base configuration for all deployable Mila language models
CLanguageNetwork
CLayerNorm	Device-templated Layer Normalization component
CLayerNormConfig
CLearningRateScheduler	Abstract base for learning-rate schedulers
CLinear	Device-templated fully connected (linear) component
CLinearConfig	Configuration object for a Linear (fully connected) layer
CLinearLRScheduler	Linear decay scheduler
CLlamaBlock
CBlockBuildContexts	The per-child build contexts this block implies
CLlamaConfig	Network-level configuration for LLaMA-style transformer networks
CLlamaModel	LLaMA 3 compatible inference model
CLlamaModelConfig	Deployment configuration for Llama language models
CLlamaTransformer	LLaMA-style transformer (decoder-only) for autoregressive token prediction
CLoss	Abstract base class for neural network loss functions
CLpe	Encoder module for token and positional embeddings (device-templated)
CLpeConfig	Configuration class for the Learned Positional Encoder
CMemoryStats	Memory allocation breakdown for a single component
CMLP	Dense feed-forward (MLP) composite component
CMLPConfig	Configuration class for the Multi-Layer Perceptron (MLP) block
CModel
CModelConfig	Abstract base configuration for all deployable Mila models
CMultiAxisPartition	Multi-axis partition for normalization over trailing dimensions
CMultiHeadAttention	Multi-Head Attention module that accepts concatenated QKV input
CMultiHeadAttentionConfig	Configuration class for Attention module
CNetwork	Root composite network container
CNetworkFactory	Factory registry for Network deserialization
COptimizer	Abstract base class for parameter optimizers
CResidual	Device-templated Residual connection component
CResidualConfig	Configuration class for Residual connection component
CRmsNorm	Device-templated RMS Normalization component
CRmsNormConfig
CRope	Device-templated RoPE component
CRopeConfig
CSampler	Abstract base for token samplers
CSamplingConfig	Model-fixed configuration for TokenSampler
CSamplingParams	Per-call sampling knobs consumed by the TokenSampler
CSerializationMetadata	Type-safe metadata container for component serialization
CSoftmax	Softmax activation module (device-templated)
CSoftmaxConfig	Configuration class for Softmax module
CSoftmaxCrossEntropy	Fused SoftmaxCrossEntropy loss module (device-templated)
CSwiglu	Gated-linear-unit (GLU-family) activation component
CSwigluConfig
CTensor	Device-aware N-dimensional tensor
CTensorBuffer	Device-agnostic buffer for storing tensor data with abstract type system
CTensorDataTypeMap	Primary template for mapping concrete C++ types to TensorDataType
CTensorDataTypeMap< __nv_fp8_e4m3 >
CTensorDataTypeMap< __nv_fp8_e5m2 >
CTensorDataTypeMap< float >	Concrete type mapping for float (FP32)
CTensorDataTypeMap< half >
CTensorDataTypeMap< nv_bfloat16 >
CTensorDataTypeMap< std::int16_t >	Concrete type mapping for 16-bit signed integer
CTensorDataTypeMap< std::int32_t >	Concrete type mapping for 32-bit signed integer
CTensorDataTypeMap< std::int8_t >	Concrete type mapping for 8-bit signed integer
CTensorDataTypeMap< std::uint16_t >	Concrete type mapping for 16-bit unsigned integer
CTensorDataTypeMap< std::uint32_t >	Concrete type mapping for 32-bit unsigned integer
CTensorDataTypeMap< std::uint8_t >	Concrete type mapping for 8-bit unsigned integer
CTensorDataTypeTraits	Compile-time traits for TensorDataType enumeration values
CTensorDataTypeTraits< TensorDataType::BF16 >	Traits specialization for 16-bit brain floating point
CTensorDataTypeTraits< TensorDataType::FP16 >	Traits specialization for 16-bit half precision floating point
CTensorDataTypeTraits< TensorDataType::FP32 >	Traits specialization for 32-bit IEEE 754 floating point
CTensorDataTypeTraits< TensorDataType::FP4_E2M1 >	Traits specialization for 4-bit floating point with E2M1 format
CTensorDataTypeTraits< TensorDataType::FP4_E3M0 >	Traits specialization for 4-bit floating point with E3M0 format
CTensorDataTypeTraits< TensorDataType::FP8_E4M3 >	Traits specialization for 8-bit floating point with E4M3 format
CTensorDataTypeTraits< TensorDataType::FP8_E5M2 >	Traits specialization for 8-bit floating point with E5M2 format
CTensorDataTypeTraits< TensorDataType::INT16 >	Traits specialization for 16-bit signed integer
CTensorDataTypeTraits< TensorDataType::INT32 >	Traits specialization for 32-bit signed integer
CTensorDataTypeTraits< TensorDataType::INT8 >	Traits specialization for 8-bit signed integer
CTensorDataTypeTraits< TensorDataType::UINT16 >	Traits specialization for 16-bit unsigned integer
CTensorDataTypeTraits< TensorDataType::UINT32 >	Traits specialization for 32-bit unsigned integer
CTensorDataTypeTraits< TensorDataType::UINT8 >	Traits specialization for 8-bit unsigned integer
CTensorHostTypeMap	Maps abstract TensorDataType to host-compatible C++ type and TensorDataType
CTensorHostTypeMap< TensorDataType::BF16 >	Host type for 16-bit brain floating point
CTensorHostTypeMap< TensorDataType::FP16 >	Host type for 16-bit half precision floating point
CTensorHostTypeMap< TensorDataType::FP32 >	Host type for 32-bit IEEE 754 floating point
CTensorHostTypeMap< TensorDataType::FP8_E4M3 >	Host type for 8-bit floating point with E4M3 format
CTensorHostTypeMap< TensorDataType::FP8_E5M2 >	Host type for 8-bit floating point with E5M2 format
CTensorHostTypeMap< TensorDataType::INT16 >	Host type for 16-bit signed integer
CTensorHostTypeMap< TensorDataType::INT32 >	Host type for 32-bit signed integer
CTensorHostTypeMap< TensorDataType::INT8 >	Host type for 8-bit signed integer
CTensorHostTypeMap< TensorDataType::UINT16 >	Host type for 16-bit unsigned integer
CTensorHostTypeMap< TensorDataType::UINT32 >	Host type for 32-bit unsigned integer
CTensorHostTypeMap< TensorDataType::UINT8 >	Host type for 8-bit unsigned integer
CTensorOps	Device-dispatched TensorOps interface template
CTensorShape	Fixed-capacity inline shape descriptor for N-dimensional tensors
CTokenEmbedding	Pure token embedding component (device-templated)
CTokenEmbeddingConfig	Configuration for the TokenEmbedding component
CTokenSampler	The standard token sampler: temperature / top-k / top-p multinomial
CUniqueIdGenerator	Thread-safe generator for unique tensor identifiers
CVulkanTensorTraits	Vulkan-specific traits for abstract tensor data types
NLogging
CConsoleSink	Thread-safe logging sink that writes formatted records to the console
CFileSink	Thread-safe logging sink that writes formatted records to a file
CLogger	Abstract logging interface and static facade
CNullSink	A logging sink that silently discards all records
NProfiling
CNvtxRange
NUtils
CStepLogger
CVersion	Semantic Version data
Nstd
Chash< Mila::Dnn::Compute::DeviceId >	Hash specialization for DeviceId