Settings

class plugins.train.model._base.settings.Settings(arguments: Namespace, mixed_precision: bool, is_predict: bool)

Bases: object

Core training settings.

Sets backend settings prior to launching the model.

Parameters:

  • arguments (Namespace) – The arguments that were passed to the train or convert process as generated from Faceswap’s command line arguments

  • mixed_precision (bool) – True if Mixed Precision training should be used otherwise False

  • is_predict (bool) – True if the model is being loaded for inference, False if the model is being loaded for training. Default: False

Attributes Summary

use_mixed_precision

True if mixed precision training has been enabled, otherwise False.

Methods Summary

check_model_precision(model, state)

Check the model's precision.

get_mixed_precision_layers(build_func, inputs)

Get and store the mixed precision layers from a full precision enabled model.

loss_scale_optimizer(optimizer)

Optimize loss scaling for mixed precision training.

Attributes Documentation

use_mixed_precision

True if mixed precision training has been enabled, otherwise False.

Methods Documentation

check_model_precision(model: keras.models.Model, state: State) → keras.models.Model

Check the model’s precision.

If this is a new model, then Rewrite an existing model’s training precision mode from mixed-float16 to float32 or vice versa.

This is not easy to do in keras, so we edit the model’s config to change the dtype policy for compatible layers. Create a new model from this config, then port the weights from the old model to the new model.

Parameters:

  • model (keras.models.Model) – The original saved keras model to rewrite the dtype

  • state (State) – The State information for the model

Return type:

The original model with the datatype updated

get_mixed_precision_layers(build_func: Callable[[list[keras.layers.Layer]], keras.models.Model], inputs: list[keras.layers.Layer]) → tuple[keras.models.Model, list[str]]

Get and store the mixed precision layers from a full precision enabled model.

Parameters:

  • build_func (Callable[[list[keras.layers.Layer]], keras.models.Model]) – The function to be called to compile the newly created model

  • inputs (list[keras.layers.Layer]) – The inputs to the model to be compiled

Returns:

  • model – The built model in fp32

  • names – The list of layer names within the full precision model that can be switched to mixed precision

Return type:

tuple[keras.models.Model, list[str]]

classmethod loss_scale_optimizer(optimizer: Optimizer) → LossScaleOptimizer

Optimize loss scaling for mixed precision training.

Parameters:

optimizer (Optimizer) – The optimizer instance to wrap

Return type:

The original optimizer with loss scaling applied

check_model_precision(model: keras.models.Model, state: State) → keras.models.Model

Check the model’s precision.

If this is a new model, then Rewrite an existing model’s training precision mode from mixed-float16 to float32 or vice versa.

This is not easy to do in keras, so we edit the model’s config to change the dtype policy for compatible layers. Create a new model from this config, then port the weights from the old model to the new model.

Parameters:

  • model (keras.models.Model) – The original saved keras model to rewrite the dtype

  • state (State) – The State information for the model

Return type:

The original model with the datatype updated

get_mixed_precision_layers(build_func: Callable[[list[keras.layers.Layer]], keras.models.Model], inputs: list[keras.layers.Layer]) → tuple[keras.models.Model, list[str]]

Get and store the mixed precision layers from a full precision enabled model.

Parameters:

  • build_func (Callable[[list[keras.layers.Layer]], keras.models.Model]) – The function to be called to compile the newly created model

  • inputs (list[keras.layers.Layer]) – The inputs to the model to be compiled

Returns:

  • model – The built model in fp32

  • names – The list of layer names within the full precision model that can be switched to mixed precision

Return type:

tuple[keras.models.Model, list[str]]

classmethod loss_scale_optimizer(optimizer: Optimizer) → LossScaleOptimizer

Optimize loss scaling for mixed precision training.

Parameters:

optimizer (Optimizer) – The optimizer instance to wrap

Return type:

The original optimizer with loss scaling applied

property use_mixed_precision: bool

True if mixed precision training has been enabled, otherwise False.