qualia_plugin_snn.learningmodel.pytorch.QuantizedSResNet module

Contains the template for a quantized residual spiking neural network.

class qualia_plugin_snn.learningmodel.pytorch.QuantizedSResNet.CreateNeuron[source]

Bases: Protocol

Signature for create_neuron from qualia_plugin_snn.learningmodel.pytorch.SNN.create_neuron().

Used to pass neuron builder to BasicBlock.

__call__(quant_params: QuantizationConfigDict) → Module[source]

Instanciate a spiking neuron.

Parameters:: quant_params (QuantizationConfigDict) – Optional quantization configuration dict in case of quantized network, see qualia_core.learningmodel.pytorch.Quantizer.Quantizer
Returns:: A spiking neuron instance
Return type:: Module

__init__(*args, **kwargs)

class qualia_plugin_snn.learningmodel.pytorch.QuantizedSResNet.QuantizedBasicBlockBuilder[source]

Bases: Protocol

Signature for basicblockbuilder.

Used to bind hyperparameters constant across all the ResNet blocks.

__call__(in_planes: int, planes: int, kernel_size: int, stride: int, padding: int) → QuantizedBasicBlock[source]

Build a BasicBlock.

Parameters:

in_planes (int) – Number of input channels
planes (int) – Number of filters (i.e., output channels) in the main branch Conv layers
kernel_size (int) – kernel_size for the main branch Conv layers
stride (int) – kernel_size for the MaxPool layers, no MaxPool layer added if 1
padding (int) – Padding for the main branch Conv layers

Returns:

A BasicBlock

Return type:

QuantizedBasicBlock

__init__(*args, **kwargs)

class qualia_plugin_snn.learningmodel.pytorch.QuantizedSResNet.QuantizedBasicBlock[source]

Bases: Module

A single quantized ResNetv1 block.

Should have topology identical to qualia_plugin_snn.learningmodel.pytorch.SResNet.BasicBlock but with layers replaced with their quantized equivalent.

Structure is:

                 |
             /        \
         |                 |
   QuantizedConv           |
         |                 |
QuantizedBatchNorm         |
         |                 |
 QuantizedMaxPool   QuantizedConv
         |                 |
    QuantizedIF  QuantizedBatchNorm
         |                 |
   QuantizedConv  QuantizedMaxPool
         |                 |
QuantizedBatchNorm   QuantizedIF
         |                 |
    QuantizedIF            |
         |                 |
             \       /
                 |
            QuantizedAdd

Main (left) branch QuantizedConv use kernel_size=kernel_size, while residual (right) branch QuantizedConv use kernel_size=1.

QuantizedBatchNorm layers will be absent if batch_norm == False

QuantizedMaxPool layer will be absent if stride == 1.

Residual (right) branch QuantizedConv layer will be asbent if in_planes==planes, except if force_projection_with_stride==True and stride != 1.

__init__(sjlayers_t: ModuleType, in_planes: int, planes: int, kernel_size: int, stride: int, padding: int, batch_norm: bool, bn_momentum: float, force_projection_with_stride: bool, create_neuron: CreateNeuron, step_mode: str, quant_params: QuantizationConfigDict) → None[source]

Construct QuantizedBasicBlock.

Parameters:

sjlayers_t (ModuleType) – Module containing the aliased quantized layers to use (1D or 2D)
in_planes (int) – Number of input channels
planes (int) – Number of filters (i.e., output channels) in the main branch QuantizedConv layers
kernel_size (int) – kernel_size for the main branch QuantizedConv layers
stride (int) – kernel_size for the QuantizedMaxPool layers, no QuantizedMaxPool layer added if 1
padding (int) – Padding for the main branch QuantizedConv layers
batch_norm (bool) – If True, add BatchNorm layer after each QuantizedConv layer
bn_momentum (float) – QuantizedBatchNorm layer momentum
force_projection_with_stride (bool) – If True, residual QuantizedConv layer is kept when stride != 1 even if in_planes == planes
create_neuron (CreateNeuron) – qualia_plugin_snn.learningmodel.pytorch.SNN.SNN.create_neuron() method to instantiate a spiking neuron
step_mode (str) – SpikingJelly step_mode from qualia_plugin_snn.learningmodel.pytorch.SNN.SNN.step_mode
quant_params (QuantizationConfigDict) – Quantization configuration dict, see qualia_core.learningmodel.pytorch.Quantizer.Quantizer

Return type:

None

expansion: int = 1: Unused

forward(input: Tensor) → Tensor[source]

Forward of quantized ResNet block.

Parameters:: input (Tensor) – Input tensor
Returns:: Output tensor
Return type:: Tensor

class qualia_plugin_snn.learningmodel.pytorch.QuantizedSResNet.QuantizedSResNet[source]

Bases: SNN

Quantized residual spiking neural network template.

Should have topology identical to qualia_plugin_snn.learningmodel.pytorch.SResNet.SResNet but with layers replaced with their quantized equivalent.

__init__(input_shape: tuple[int, ...], output_shape: tuple[int, ...], filters: list[int], kernel_sizes: list[int], num_blocks: list[int], strides: list[int], paddings: list[int], quant_params: QuantizationConfig, prepool: int = 1, postpool: str = 'max', batch_norm: bool = False, bn_momentum: float = 0.1, force_projection_with_stride: bool = True, neuron: RecursiveConfigDict | None = None, timesteps: int = 2, dims: int = 1, basicblockbuilder: QuantizedBasicBlockBuilder | None = None) → None[source]

Construct QuantizedSResNet.

Structure is:

  QuantizedInput
         |
 QuantizedAvgPool
         |
   QuantizedConv
         |
QuantizedBatchNorm
         |
    QuantizedIF
         |
QuantizedBasicBlock
         |
         …
         |
QuantizedBasicBlock
         |
QuantizedGlobalPool
         |
      Flatten
         |
  QuantizedLinear

Parameters:

input_shape (tuple[int, ...]) – Input shape
output_shape (tuple[int, ...]) – Output shape
filters (list[int]) – List of out_channels for QuantizedConv layers inside each QuantizedBasicBlock group, must be of the same size as num_blocks, first element is for the first QuantizedConv layer at the beginning of the network
kernel_sizes (list[int]) – List of kernel_size for QuantizedConv layers inside each QuantizedBasicBlock group, must of the same size as num_blocks, first element is for the first QuantizedConv layer at the beginning of the network
num_blocks (list[int]) – List of number of QuantizedBasicBlock in each group, also defines the number of QuantizedBasicBlock groups inside the network
strides (list[int]) – List of kernel_size for QuantizedMaxPool layers inside each QuantizedBasicBlock group, must of the same size as num_blocks, stride is applied only to the first QuantizedBasicBlock of the group, next QuantizedBasicBlock in the group use a stride of 1, first element is the stride of the first QuantizedConv layer at the beginning of the network
paddings (list[int]) – List of padding for QuantizedConv layer inside each QuantizedBasicBlock group, must of the same size as num_blocks, first element is for the first QuantizedConv layer at the beginning of the network
prepool (int) – QuantizedAvgPool layer kernel_size to add at the beginning of the network, no layer added if 0
postpool (str) – Quantized global pooling layer type after all QuantizedBasicBlock, either max for QuantizedMaxPool or avg for QuantizedAvgPool
batch_norm (bool) – If True, add a QuantizedBatchNorm layer after each QuantizedConv layer, otherwise no layer added
bn_momentum (float) – QuantizedBatchNorm momentum
force_projection_with_stride (bool) – If True, residual QuantizedConv layer is kept when stride != 1 even if in_planes == planes inside a QuantizedBasicBlock
neuron (RecursiveConfigDict | None) – Spiking neuron configuration, see qualia_plugin_snn.learningmodel.pytorch.SNN.SNN.__init__()
timesteps (int) – Number of timesteps
dims (int) – Either 1 or 2 for 1D or 2D convolutional network.
basicblockbuilder (QuantizedBasicBlockBuilder | None) – Optional function with QuantizedBasicBlockBuilder.__call__() signature to build a basic block after binding constants common across all basic blocks
quant_params (QuantizationConfig)

Return type:

None

forward(input: Tensor) → Tensor[source]

Parameters:: input (Tensor)
Return type:: Tensor