Hooks

Inference

XPM Configxpmir.context.Hook

Bases: Config

Base class for all hooks

XPM Configxpmir.context.InitializationHook

Bases: Hook

Base class for hooks before/after initialization

after(context: Context)

Called after initialization

before(context: Context)

Called before initialization

Learning

Hooks can be used to modify the learning process

XPM Configxpmir.learning.context.TrainingHook

Bases: Config

Base class for all training hooks

XPM Configxpmir.learning.context.InitializationTrainingHook

Bases: InitializationHook

Base class for hooks called at each epoch (before/after)

after(state: TrainerContext)

Called after initialization

before(state: TrainerContext)

Called before initialization

XPM Configxpmir.learning.context.StepTrainingHook

Bases: TrainingHook

Base class for hooks called at each step (before/after)

after(state: TrainerContext)

Called after a training step

before(state: TrainerContext)

Called before a training step

Distributed

Hooks can be used to distribute a model over GPUs

XPM Configxpmir.distributed.DistributableModel

Bases: Config

A model that can be distributed over GPUs

Subclasses must implement distribute_models()

distribute_models(update: Callable[[TorchModule], TorchModule])

This method is called with an update parameter that should be used to update all the torch modules that we need to distribute on GPUs

XPM Configxpmir.distributed.DistributedHook(*, models)

Bases: InitializationHook

Hook to distribute the model processing

When in multiprocessing/multidevice, use torch.nn.parallel.DistributedDataParallel ,otherwise use torch.nn.DataParallel.

models: List[xpmir.distributed.DistributableModel]

The model to distribute over GPUs