Mila
Deep Neural Network Library
Loading...
Searching...
No Matches
File List
Here is a list of all documented files with brief descriptions:
 Mila
 Src
 Data
 Core
 
FileHeader.ixx		Common file header structure for Mila data files
 
TokenizerTrainer.ixx		Abstract trainer interface for building tokenizers' vocabularies
 
TrainerFactory.ixx		Factory helpers to construct tokenizer trainers and load vocabularies
 Loaders
 
DataLoader.ixx		Device-agnostic data loader interface using abstract tensor data types
 Tokenizers
 Bpe
 
BpePreTokenizationMode.ixx		Pre-tokenization mode enumeration for GPT-4 style BPE tokenizers
 
BpeTokenizer.ixx		Unified BPE tokenizer for GPT-2, Llama 3.x, Mistral, and Gemma model families
 
BpeTrainer.ixx		BPE vocabulary trainer with incremental corpus accumulation
 
BpeVocabulary.ixx		BPE vocabulary for GPT-2, Llama 3.x, and Mistral model families
 
BpeVocabularyConfig.ixx		Unified configuration for BPE vocabulary construction and runtime properties
 Char
 
CharTokenizer.ixx		Character-level tokenizer implementing the Tokenizer API
 
CharTrainer.ixx		Character-level tokenizer trainer for corpus accumulation and vocabulary building
 
CharVocabulary.ixx		Character vocabulary with factory-based construction
 
CharVocabularyConfig.ixx		Configuration for Character-level tokenizer training
 
SpecialTokens.ixx		Configuration for special tokens used across all tokenizer types
 
TokenizerVocabulary.ixx		Abstract interface for tokenizer vocabularies used by data pipelines
 Distribution
 
CurlHttpTransport.ixx		Libcurl as an IHttpTransport: one request, exactly the headers given
 
Environment.ixx		Reading an environment variable, once, safely
 
HttpClient.ixx		How to ask, and how to read the answer: redirects, the token rule, resume, status
 
HttpTransport.ixx		One HTTP request, moved
 
HttpTransportBackend.Curl.ixx		The HTTP transport this build has: libcurl
 
HttpTransportBackend.Null.ixx		The transport for a build with no HTTP client of its own: one that refuses, by name
 
HuggingFaceHub.ixx		The HuggingFace implementation of IModelHub, and its token discovery
 
ModelCoordinate.ixx		How a hub repository is addressed: [hf:]<owner>/<repository>[@<revision>]
 
ModelHub.ixx		The abstraction over a non-local model hub
 
ModelManifest.ixx		The mila.json schema: what files compose a model, and for which Mila
 
ModelPackage.ixx		A package directory: the manifest, the files it declares, and whether they agree
 
ModelResolver.ixx		Pulls a published model into the store: manifest, variant selection, blobs, record
 
ModelStore.ixx		The local store of installed models: content-addressed blobs plus the records that name them
 
Sha256.ixx		Streaming SHA-256, for verifying downloaded model blobs against their digests
 Dnn
 Components
 Activations
 Activation
 
Activation.Config.ixx		Configuration for the unified elementwise Activation component
 
Activation.ixx		Unified elementwise activation component
 Gelu
 
Gelu.Config.ixx		Configuration for the GELU activation module
 
Gelu.ixx		GELU activation component implementation
 
ActivationType.ixx		Definition of activation function types used throughout the Mila library
 
ApproximationMethod.ixx		Shared approximation method enum for activation functions
 Attention
 GQA
 
GroupedQueryAttention.Config.ixx		Configuration interface for the Grouped-Query Attention component
 
GroupedQueryAttention.ixx		Grouped-Query Attention module (concatenated QKV input)
 MHA
 
MultiHeadAttention.Config.ixx		Configuration interface for the Attention module
 
MultiHeadAttention.ixx		Multi-Head Attention module (concatenated QKV input)
 
AttentionType.ixx		Defines attention mechanism types used by transformer components
 Connections
 
ConnectionType.ixx		Definition of connection function types used by the Mila DNN library
 
Residual.ixx		Device-templated Residual connection component
 
ResidualConfig.ixx		Configuration for the Residual component
 Embeddings
 
TokenEmbedding.Config.ixx		Configuration for the TokenEmbedding component
 
TokenEmbedding.ixx		Device-templated TokenEmbedding component
 Encodings
 Lpe
 
Lpe.Config.ixx		Configuration interface for the Encoder module in the Mila DNN framework
 
Lpe.ixx		Device-templated Encoder module for token and positional embeddings
 Rope
 
Rope.Config.ixx		Configuration for Rotary Position Embedding (RoPE) component
 
Rope.ixx		Rotary positional embedding (RoPE) component
 
EncodingType.ixx		Positional encoding strategy selection used by Transformer components
 FFN
 GatedMLP
 
GatedMLP.Config.ixx		Configuration for the gated feed-forward (GatedMLP) block
 
GatedMLP.ixx		Gated feed-forward (GatedMLP) block: fused gate+up -> Swiglu gate -> down
 MLP
 
MLP.Config.ixx		Configuration for the MLP block
 
MLP.Dispatch.ixx		Activation dispatch helpers for MLP
 
MLP.ixx		Dense feed-forward (MLP) block: Linear -> GELU -> Linear
 Swiglu
 
Swiglu.Config.ixx		Configuration for the SwiGLU activation component
 
Swiglu.ixx		SwiGLU activation component implementation
 Linear
 
Linear.ixx		Device-templated Linear (fully connected) component
 
LinearConfig.ixx		Configuration for the Linear (fully connected) layer
 Losses
 
CrossEntropyConfig.ixx		Configuration for the fused SoftmaxCrossEntropy loss module
 
SoftmaxCrossEntropy.ixx		Device-templated fused SoftmaxCrossEntropy loss module
 Normalization
 LayerNorm
 
LayerNorm.Config.ixx		Configuration for Layer Normalization component
 
LayerNorm.ixx		Layer Normalization component
 RmsNorm
 
RmsNorm.Config.ixx		Configuration for RMS Normalization component
 
RmsNorm.ixx		RMS Normalization component
 
NormType.ixx		Normalization layer type enumeration used by Transformer components
 
Softmax.ixx		Device-templated Softmax activation module
 
SoftmaxConfig.ixx		Configuration interface for the Softmax module in the Mila DNN framework
 Transformers
 Gemma
 
Gemma.Block.ixx		Gemma 4 decoder block (inference: prefill + decode), one type per attention kind
 
Gemma.Config.ixx		Network-level configuration for Gemma 4 transformer networks
 
Gemma.ixx		Gemma 4 decoder-only transformer network (inference: prefill + decode)
 
Gemma.Protocol.ixx		Canonical Gemma 4 native token grammar: parse/format for the model's registered turn / channel / tool-call / tool-response vocabulary
 
IDecoderLayer.ixx		Virtual inference interface for a heterogeneous decoder layer list
 Gpt
 
Gpt.Config.ixx		Network-level configuration for GPT-style transformer networks
 
GptBlock.Config.ixx		Configuration for GPT-style transformer block (block-level)
 
GptBlock.ixx		Transformer encoder block implementation
 
GptTransformer.ixx		GPT-2 style transformer network (decoder-only) for autoregressive language modeling
 LlaMa
 
Llama.Block.ixx		LLaMA transformer block – module partition of LlamaTransformer
 
Llama.Config.ixx		LLaMA network-level configuration
 
Llama.ixx		LLaMA-style decoder-only transformer network
 
GenerateParams.ixx		Per-call generation request: loop control plus the sampling shape
 
SamplingParams.ixx		Per-call sampling parameters – the "sampling shape"
 
MilaComponents.ixx		Aggregate module that re-exports Mila built-in DNN components
 Compute
 Devices
 Cpu
 Tensors
 
CpuTensorDataTypeTraits.ixx		CPU-specific tensor trait specializations
 
CpuDevice.ixx		Implementation of CPU-based compute device for the Mila framework
 
CpuDeviceTypeTraits.ixx		DeviceTypeTraits specialization for the CPU device
 
CpuExecutionContext.ixx		CPU-specific execution context specialization
 
CpuMemoryResourceTraits.ixx		CPU-specific memory resource traits and specializations
 Cuda
 Profiling
 
CudaTimer.ixx		GPU-accurate interval timer using a CUDA event pair
 Tensors
 
CudaTensorDataType-Maps.ixx		CUDA-specific mappings between abstract TensorDataType and concrete CUDA native types
 
CudaTensorDataType-Specializations.ixx		CUDA concrete type trait specializations
 
CudaTensorDataType.ixx		CUDA-specific tensor data type trait system - Primary module interface
 
CudaTensorDataTypes-CublasLtTypes.ixx		Compile-time mapping from abstract TensorDataType to cuBLASLt runtime cudaDataType_t enums
 
CudaDevice.ixx		Implementation of CUDA-based compute device for the Mila framework
 
CudaDeviceProps.ixx		CUDA device properties wrapper with caching and convenience methods
 
CudaDeviceTypeTraits.ixx		DeviceTypeTraits specialization for CUDA devices
 
CudaExecutionContext.ixx		CUDA-specific execution context specialization
 
CudaMemoryResourceTraits.ixx		CUDA-specific memory resource traits and specializations
 Metal
 Tensors
 
MetalTensorTraits.ixx		Metal-specific tensor trait specializations
 
MetalDevice.ixx		Implementation of Metal-based compute device for the Mila framework
 
MetalDevicePlugin.ixx		Metal device plugin for device-agnostic registration and discovery
 
MetalExecutionContext.ixx		Metal-specific execution context specialization
 
MetalMemoryResource.ixx		Metal-specific memory resource implementation for Apple GPU compute
 Rocm
 Tensors
 
RocmTensorTraits.ixx		Vulkan-specific tensor trait specializations for explicit graphics compute
 
RocmDevice.ixx		Minimal Vulkan-based compute device for the Mila framework
 
RocmExecutionContext.ixx		Vulkan-specific execution context specialization
 
RocmMemoryResource.ixx		Vulkan memory resource implementation for cross-platform GPU compute
 Operations
 
GqaState.ixx		Non-owning transient scratch state for CudaGqaOp inference paths
 
IKVCacheLifecycle.ixx		Interface for operations that own and manage a KV cache
 
IKvInference.ixx		KV-cache compute interface for modern attention backends (GQA and beyond)
 
IPackedKvInference.ixx		KV-cache inference interface for GPT-style packed-QKV MHA backends
 
IPositionalDecode.ixx		Interface for unary operations whose output depends on absolute token position
 
IPositionalPairedOp.ixx		Interface for paired operations whose output depends on absolute token position
 
OperationBase.ixx		Core abstraction for neural network operations in the Mila framework
 
OperationTraits.ixx		Aggregator for the unified operation traits dispatch table
 
OperationTraits.Template.ixx		Unified compile-time dispatch template mapping (OperationType, DeviceType, TPrecision, TPolicy) to a concrete operation type
 
OperationType.ixx		Defines the operation types supported by the compute framework
 Registry
 
DeviceRegistrar.ixx		Device-agnostic registrar for automatic device discovery and registration
 
DeviceRegistry.ixx		Central registry for discovered compute devices
 
DeviceRegistryHelpers.ixx		Utility functions for compute device discovery and management
 
Device.ixx		Abstract compute device interface and device identifier factory
 
DeviceId.ixx		Lightweight device identifier value type
 
DeviceType.ixx		Device type definitions and conversion utilities for compute devices
 
ExecutionContext.ixx		Templated execution context framework for compute operations and stream management
 
IExecutionContext.ixx		Minimal type-erased execution context interface
 
MemoryResource.ixx		Defines a clean memory resource abstraction focused on allocation responsibilities
 
MemoryResourceTraits.ixx		Compute backend memory resource traits for dispatch optimization
 Core
 
Component.BuildContext.ixx		Build-time context passed to Component::build()
 
Component.ixx		Base component interface for Mila DNN components
 
Component.MemoryStats.ixx		Memory allocation statistics for component inspection
 
Component.TrainingMode.ixx		Runtime behavioral mode for training-built Components
 
ComponentConfig.ixx		Base configuration interface for DNN components
 
ComponentFactory.ixx		Factory helpers for reconstructing built-in components from archives
 
ComponentType.ixx		Enumeration of built-in component types supported by the deserializer
 
CompositeComponent.ixx		Abstract container for managing child components
 
Decoder.ixx		Base interface for Mila decoders
 
LanguageModel.ixx		Abstract base for Mila autoregressive language models
 
LanguageModelConfig.ixx		CRTP base configuration for all deployable Mila language models
 
LanguageNetwork.ixx		Abstract base for language model networks
 
LearningRateScheduler.ixx		Learning-rate scheduler base and common concrete schedules
 
Model.ixx		Abstract base class for all Mila models
 
Model.RuntimeMode.ixx		Build-time execution mode for component buffer allocation policy
 
ModelConfig.ixx		Base configuration for all deployable Mila models
 
ModelType.ixx		Architecture identity for top-level model networks
 
Network.ixx		Root composite network container
 
OptimizerBase.ixx		Base interface for neural network parameter optimizers
 
SamplerBase.ixx		Base interface for Mila token samplers
 
TokenStreamer.ixx		Token streaming abstractions for autoregressive generation
 Models
 
GemmaModel.ixx		Gemma 4 inference model
 
GemmaModelConfig.ixx		Deployment configuration for Gemma language models
 
GptModel.ixx		GPT inference model
 
GptModelConfig.ixx		Deployment configuration for Gpt2 language models
 
LlamaModel.ixx		LLaMA inference model
 
LlamaModelConfig.ixx		Deployment configuration for Llama language models
 
QuantizationDispatch.ixx		The one place a runtime quantization setting becomes a compile-time policy
 Optimizers
 
AdamW.ixx		AdamW optimizer wrapper using fluent AdamWConfig
 
AdamWConfig.ixx		AdamW optimizer configuration
 Quantization
 KvCache
 
Policy.ixx		KV cache compression policy concept and identity struct
 
QuantPolicy.ixx		Quantization-specific KV cache compression policies
 
Quantization.ixx		Umbrella module for the Mila quantization subsystem
 Samplers
 
SamplingConfig.ixx		Construction-time configuration for the token sampler
 
TokenSampler.ixx		Device-agnostic token sampler facade
 Serialization
 
ArchiveSerializer.ixx		Interface for hierarchical archive formats (ZIP, tar, etc.)
 
ModelArchive.ixx		Structured archive helper used by component save/load implementations
 
OpenMode.ixx		Archive open mode for model serializers
 
PretrainedReader.ixx		Reader for Mila pretrained binary format
 
SafeTensors.ixx		Safetensors container support: dtype naming and a two-phase writer
 
SerializationMetadata.ixx		Type-safe metadata container for component serialization
 
Serializer.ixx		Minimal base interface for all serialization backends
 
ZipSerializer.ixx		ZIP-based ModelSerializer implementation using miniz
 Tensors
 Operations
 
TensorOps-Base.ixx		Base declaration for device-specific TensorOps specializations
 
TensorOps.Fill.ixx		High-level initializer helpers (device-dispatching) for tensors
 
TensorOps.Math.ixx		Device-dispatching math helpers for tensor arithmetic operations
 
TensorOps.Random.ixx		Device-dispatching random initialization for tensors
 
TensorOps.Structural.ixx		Device-dispatched structural operations for tensors
 
TensorOps.Transfer.ixx		Tensor transfer utilities – copy/dispatch helpers for tensor data movement
 
TensorOps.Zero.ixx		Device-dispatched fast zero operation for tensor buffers
 
ITensor.ixx		Interface providing minimal representation for tensor data across different implementations
 
Tensor.ixx		Device-aware tensor type with scalar support
 
Tensor.Serialization.ixx		Tensor-specific serialization helpers and metadata
 
Tensor.Types.ixx		Core shape, stride, and index types for the Mila tensor API
 
TensorBuffer.ixx		Device-agnostic memory management layer for tensor data using abstract data types
 
TensorDataType.ixx		Abstract tensor data type enumeration and traits system for device-agnostic tensor operations
 
TensorDataTypeMap.ixx		Concrete C++ type to abstract TensorDataType mapping utilities
 
TensorDataTypeTraits.ixx		Compile-time traits for the abstract TensorDataType enumeration
 
TensorHostTypeMap.ixx		Device-agnostic host type mapping for abstract TensorDataType enumeration
 Logging
 
ConsoleSink.ixx		Console-based logging sink for the Mila logging infrastructure
 
FileSink.ixx		File-based logging sink for the Mila logging infrastructure
 
Logger.ixx		Abstract logging interface and static facade for the Mila logging infrastructure
 
NullSink.ixx		No-op logging sink for the Mila logging infrastructure
 Utils
 
RandomGenerator.ixx		Provides a centralized random number generator for the Mila library
 
Mila.ixx		Mila public API umbrella module - the single supported entry point (import Mila;)
 
Version.ixx		Semantic version type and Mila library version constants