Module module 

Core module infrastructure and neural-network initializers.

Structs

Attribute

Attribute to print in the display method.

Content

Struct to store the attributes of a module for formatting.

DisplaySettings

Custom module display settings.

Flag

A boolean parameter value used for module-owned control state.

IgnoredDeprecated

Container to satisfy the Module trait for types that are not modules.

Lora

A Reparameterizer that attaches LoRA adapters to 2-D weight parameters.

LoraAdapter

A LoRA (Low-Rank Adaptation) adapter attached to a frozen weight parameter.

Param

Parameters are identified, traversable values owned by modules.

ParamGroup

A way to represent a group of parameter for a Burn module.

ParamId

Unique ID for a parameter of a module.

ParamMapper

Applies transformations when loading and saving parameters.

QLora

A Reparameterizer implementing QLoRA: it quantizes the (frozen) base weights and attaches full-precision trainable LoRA adapters to 2-D weights.

Quantizer

Describes how to quantize a module.

RunningState

A state that can be updated during the forward pass while being thread safe.

Enums

Initializer

Enum specifying with what values a tensor should be initialized

ParamGroupError

Errors tied to ParamGroup’s.

Traits

Module

Trait for all neural network modules.

ModuleDisplay

Trait to implement custom display settings for a module.

ModuleDisplayDefault

Default display settings for a module.

ModuleMapper

Module mapper trait for transforming module parameters.

ModuleVisitor

Module visitor trait for traversing and inspecting module parameters.

Parameter

A tensor-backed ParameterValue with autodiff, device and loading capabilities.

ParameterValue

A value that can be identified and traversed inside a Param.

Reparameterization

A rank-specific parameter reparameterization.

Reparameterizer

Defines how floating-point parameters are prepared for reparameterization.

Functions

extract_type_name

Extracts the short name of a type T

list_param_ids

Lists the IDs of all tensor-backed parameters in a module.

Type Aliases

Devices

Type alias to Vec<Device> which supports no_std environments, but automatically using the alloc crate.

Derive Macros

Module

Derive macro for the Module trait.