Layer API¶

The Layer class represents a basic computational unit in MLLM’s neural network framework. Layers are typically used to implement specific operations like linear transformations, convolutions, or activation functions.

#include "mllm/nn/Layer.hpp"

Base Class¶

class Layer¶: Base class for neural network layers. Layers are typically used to implement specific operations.

Constructors¶

explicit Layer::Layer(const LayerImpl::ptr_t &impl)¶

Constructor with LayerImpl pointer.

Parameters:: impl – Shared pointer to LayerImpl instance

template<typename T> Layer::Layer(OpTypes op_type, const T &cargo)¶

Constructor with operation type and options.

Parameters:

op_type – Operation type for the layer
cargo – Options for the operation

Core Methods¶

LayerImpl::ptr_t Layer::impl() const¶

Get the underlying LayerImpl pointer.

Returns:: Shared pointer to LayerImpl

std::vector<Tensor> Layer::__main(const std::vector<Tensor> &inputs)¶

Main execution method for the layer.

Parameters:: inputs – Input tensors
Returns:: Output tensors

OpTypes Layer::opType() const¶

Get the operation type of the layer.

Returns:: Operation type

BaseOpOptionsBase &Layer::refOptions()¶

Get reference to the layer’s options.

Returns:: Reference to BaseOpOptionsBase

Layer &Layer::to(DeviceTypes device_type)¶

Move the layer to specified device.

Parameters:: device_type – Target device type (kCPU, kCUDA, etc.)
Returns:: Reference to this layer

void Layer::__fmt_print(std::stringstream &ss)¶

Format print information about the layer.

Parameters:: ss – String stream to write formatted output

Helper Macros¶

MLLM_LAYER_ANY_INPUTS_1_OUTPUTS_FORWARD¶: Macro for defining operator() with any number of inputs and 1 output.

MLLM_LAYER_ANY_INPUTS_2_OUTPUTS_FORWARD¶: Macro for defining operator() with any number of inputs and 2 outputs.

MLLM_LAYER_ANY_INPUTS_3_OUTPUTS_FORWARD¶: Macro for defining operator() with any number of inputs and 3 outputs.