nncf.common.pruning.schedulers

Classes

PruningScheduler

This is the class from which all pruning schedulers inherit.

class nncf.common.pruning.schedulers.PruningScheduler(controller, params)

Bases: nncf.common.schedulers.BaseCompressionScheduler

This is the class from which all pruning schedulers inherit.

A pruning scheduler is an object which specifies the pruning level at each training epoch. It involves a scheduling algorithm, defined in the _calculate_pruning_level() method and a state (some parameters required for current pruning level calculation) defined in the __init__() method.

Parameters:

• controller (nncf.api.compression.CompressionAlgorithmController) – Pruning algorithm controller.

• params (dict) – Parameters of the scheduler in the JSON-like dictionary form. Passed as-is from the corresponding section of the NNCF config file .json section (https://openvinotoolkit.github.io/nncf/schema).

property current_pruning_level: float

Returns pruning level for the current_epoch.

Returns:

Current sparsity level.

Return type:

float

epoch_step(next_epoch=None)

Should be called at the beginning of each training epoch to prepare the pruning method to continue training the model in the next_epoch.

Parameters:

next_epoch (Optional[int]) – The epoch index for which the pruning scheduler will update the state of the pruning method.

Return type:

None

step(next_step=None)

Should be called at the beginning of each training step to prepare the pruning method to continue training the model in the next_step.

Parameters:

next_step (Optional[int]) – The global step index for which the pruning scheduler will update the state of the pruning method.

Return type:

None