aihwkit.cloud.converter.v1.i_mappings module

Mappings for version 1 of the AIHW Composer format.

class aihwkit.cloud.converter.v1.i_mappings.Function(id_, args)

Bases: object

Mapping for a function-like entity.

Parameters:

• id_ (str) –

• args (Dict) –

from_proto(source, cls, default=None)

Convert a proto object into a destination object.

Parameters:

• source (Any) –

• cls (type) –

• default (Any | None) –

Return type:

object

get_argument_from_proto(source, field, default=None)

Get the value of an argument.

Parameters:

• source (Any) –

• field (str) –

• default (Any | None) –

Return type:

Dict

get_field_value_to_proto(source, field, default=None)

Get the value of a field.

Parameters:

• source (Any) –

• field (str) –

• default (Any | None) –

Return type:

Any

to_proto(source, proto_cls)

Convert a source object into a destination object.

Parameters:

• source (object) –

• proto_cls (type) –

Return type:

object

class aihwkit.cloud.converter.v1.i_mappings.InverseMappings

Bases: object

Mappings between AIHW Composer format and Python entities.

activation_functions = {'LeakyReLU': <class 'torch.nn.modules.activation.LeakyReLU'>, 'LogSigmoid': <class 'torch.nn.modules.activation.LogSigmoid'>, 'LogSoftmax': <class 'torch.nn.modules.activation.LogSoftmax'>, 'ReLU': <class 'torch.nn.modules.activation.ReLU'>, 'Sigmoid': <class 'torch.nn.modules.activation.Sigmoid'>, 'Softmax': <class 'torch.nn.modules.activation.Softmax'>, 'Tanh': <class 'torch.nn.modules.activation.Tanh'>}

datasets = {'fashion_mnist': <class 'torchvision.datasets.mnist.FashionMNIST'>, 'svhn': <class 'torchvision.datasets.svhn.SVHN'>}

layers = {'AnalogConv2d': <class 'aihwkit.nn.modules.conv.AnalogConv2d'>, 'AnalogConv2dMapped': <class 'aihwkit.nn.modules.conv_mapped.AnalogConv2dMapped'>, 'AnalogLinear': <class 'aihwkit.nn.modules.linear.AnalogLinear'>, 'AnalogLinearMapped': <class 'aihwkit.nn.modules.linear_mapped.AnalogLinearMapped'>, 'BatchNorm2d': <class 'torch.nn.modules.batchnorm.BatchNorm2d'>, 'Conv2d': <class 'torch.nn.modules.conv.Conv2d'>, 'ConvTranspose2d': <class 'torch.nn.modules.conv.ConvTranspose2d'>, 'Flatten': <class 'torch.nn.modules.flatten.Flatten'>, 'Linear': <class 'torch.nn.modules.linear.Linear'>, 'MaxPool2d': <class 'torch.nn.modules.pooling.MaxPool2d'>}

loss_functions = {'BCELoss': <class 'torch.nn.modules.loss.BCELoss'>, 'CrossEntropyLoss': <class 'torch.nn.modules.loss.CrossEntropyLoss'>, 'MSELoss': <class 'torch.nn.modules.loss.MSELoss'>, 'NLLLoss': <class 'torch.nn.modules.loss.NLLLoss'>}

optimizers = {'AnalogSGD': <class 'aihwkit.optim.analog_optimizer.AnalogSGD'>}

presets = {'InferenceRPUConfig': <class 'aihwkit.simulator.configs.configs.InferenceRPUConfig'>, 'OldWebComposerInferenceRPUConfig': <class 'aihwkit.simulator.presets.web.OldWebComposerInferenceRPUConfig'>, 'WebComposerInferenceRPUConfig': <class 'aihwkit.simulator.presets.web.WebComposerInferenceRPUConfig'>}

class aihwkit.cloud.converter.v1.i_mappings.LayerFunction(id_, args)

Bases: Function

Mapping for a function-like entity (Layer).

Parameters:

• id_ (str) –

• args (Dict) –

get_argument_from_proto(source, field, default=None)

Get the value of an argument.

Raises ConversionError

Parameters:

• source (Any) –

• field (str) –

• default (Any | None) –

Return type:

Dict

get_field_value_to_proto(source, field, default=None)

Get the value of a field.

Raises ConversionError

Parameters:

• source (Any) –

• field (str) –

• default (Any | None) –

Return type:

Any

class aihwkit.cloud.converter.v1.i_mappings.Mappings

Bases: object

Mappings between Python entities and AIHW format.

activation_functions = {<class 'torch.nn.modules.activation.LeakyReLU'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>, <class 'torch.nn.modules.activation.LogSigmoid'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>, <class 'torch.nn.modules.activation.LogSoftmax'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>, <class 'torch.nn.modules.activation.ReLU'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>, <class 'torch.nn.modules.activation.Sigmoid'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>, <class 'torch.nn.modules.activation.Softmax'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>, <class 'torch.nn.modules.activation.Tanh'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>}

datasets = {<class 'torchvision.datasets.mnist.FashionMNIST'>: 'fashion_mnist', <class 'torchvision.datasets.svhn.SVHN'>: 'svhn'}

layers = {<class 'aihwkit.nn.modules.conv.AnalogConv2d'>: <aihwkit.cloud.converter.v1.i_mappings.LayerFunction object>, <class 'aihwkit.nn.modules.conv_mapped.AnalogConv2dMapped'>: <aihwkit.cloud.converter.v1.i_mappings.LayerFunction object>, <class 'aihwkit.nn.modules.linear.AnalogLinear'>: <aihwkit.cloud.converter.v1.i_mappings.LayerFunction object>, <class 'aihwkit.nn.modules.linear_mapped.AnalogLinearMapped'>: <aihwkit.cloud.converter.v1.i_mappings.LayerFunction object>, <class 'torch.nn.modules.batchnorm.BatchNorm2d'>: <aihwkit.cloud.converter.v1.i_mappings.LayerFunction object>, <class 'torch.nn.modules.conv.Conv2d'>: <aihwkit.cloud.converter.v1.i_mappings.LayerFunction object>, <class 'torch.nn.modules.conv.ConvTranspose2d'>: <aihwkit.cloud.converter.v1.i_mappings.LayerFunction object>, <class 'torch.nn.modules.flatten.Flatten'>: <aihwkit.cloud.converter.v1.i_mappings.LayerFunction object>, <class 'torch.nn.modules.linear.Linear'>: <aihwkit.cloud.converter.v1.i_mappings.LayerFunction object>, <class 'torch.nn.modules.pooling.MaxPool2d'>: <aihwkit.cloud.converter.v1.i_mappings.LayerFunction object>}

loss_functions = {<class 'torch.nn.modules.loss.BCELoss'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>, <class 'torch.nn.modules.loss.CrossEntropyLoss'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>, <class 'torch.nn.modules.loss.MSELoss'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>, <class 'torch.nn.modules.loss.NLLLoss'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>}

optimizers = {<class 'aihwkit.optim.analog_optimizer.AnalogSGD'>: <aihwkit.cloud.converter.v1.i_mappings.Function object>}

presets = {<class 'aihwkit.simulator.configs.configs.InferenceRPUConfig'>: 'InferenceRPUConfig', <class 'aihwkit.simulator.presets.web.OldWebComposerInferenceRPUConfig'>: 'OldWebComposerInferenceRPUConfig', <class 'aihwkit.simulator.presets.web.WebComposerInferenceRPUConfig'>: 'WebComposerInferenceRPUConfig'}

class aihwkit.cloud.converter.v1.i_mappings.Type(attribute_type, field, fn)

Bases: tuple

attribute_type

Alias for field number 0

field

Alias for field number 1

fn

Alias for field number 2

aihwkit.cloud.converter.v1.i_mappings.build_inverse_mapping(mapping)

Create the inverse mapping between Python entities and AIHW Composer formats.

Parameters:

mapping (Dict) –

Return type:

Dict