Channel pruning

Context

Channel pruning (CP) is a model compression technique that removes less-important input channels from layers. AIMET supports channel pruning of 2D convolution (Conv2D) layers.

Procedure

The following figure illustrates the different steps in channel pruning a layer. These steps are repeated for all layers selected for compression in order of occurrence from the top of the model.

The steps are explained below.

Channel selection

For a layer and a specified compression ratio, channel selection analyzes the magnitude of each input channel based its kernel weights. It chooses the lowest-magnitude channels to be pruned.

Winnowing

Winnowing is the process of removing the input channels identified in channel selection, resulting in compressed tensors:

Once one or more input channels of a layer are removed, corresponding output channels of an upstream layer are also be removed to gain further compression. Skip-connections or residuals sometimes prevent upstream layers from being output-pruned.

For more details on winnowing, see Winnowing.

Weight reconstruction

The final step in CP is to adjust the weight and bias parameters of a pruned layer to try and match pre-pruning output values. AIMET does this by performing linear regression on random samples of the layer’s input from the pruned model against corresponding output from the original model.

Workflow

Code example

Setup

PyTorch

Compression using Channel Pruning

PyTorch

Compressing using Channel Pruning in auto mode.

Compressing using Channel Pruning in manual mode.

API

PyTorch

Top-level API for Compression

class aimet_torch.compress.ModelCompressor

AIMET model compressor: Enables model compression using various schemes

static ModelCompressor.compress_model(model, eval_callback, eval_iterations, input_shape, compress_scheme, cost_metric, parameters, trainer=None, visualization_url=None)

Compress a given model using the specified parameters

Parameters:

• model (Module) – Model to compress

• eval_callback (Callable[[Any, Optional[int], bool], float]) – Evaluation callback. Expected signature is evaluate(model, iterations, use_cuda). Expected to return an accuracy metric.

• eval_iterations – Iterations to run evaluation for

• trainer – Training Class: Contains a callable, train_model, which takes model, layer which is being fine tuned and an optional parameter train_flag as a parameter None: If per layer fine tuning is not required while creating the final compressed model

• input_shape (Tuple) – Shape of the input tensor for model

• compress_scheme (CompressionScheme) – Compression scheme. See the enum for allowed values

• cost_metric (CostMetric) – Cost metric to use for the compression-ratio (either mac or memory)

• parameters (Union[SpatialSvdParameters, WeightSvdParameters, ChannelPruningParameters]) – Compression parameters specific to given compression scheme

• visualization_url – url the user will need to input where visualizations will appear

Return type:

Tuple[Module, CompressionStats]

Returns:

A tuple of the compressed model, and compression statistics

Greedy Selection Parameters

class aimet_torch.common.defs.GreedySelectionParameters(target_comp_ratio, num_comp_ratio_candidates=10, use_monotonic_fit=False, saved_eval_scores_dict=None)

Configuration parameters for the Greedy compression-ratio selection algorithm

Variables:

• target_comp_ratio – Target compression ratio. Expressed as value between 0 and 1. Compression ratio is the ratio of cost of compressed model to cost of the original model.

• num_comp_ratio_candidates – Number of comp-ratio candidates to analyze per-layer More candidates allows more granular distribution of compression at the cost of increased run-time during analysis. Default value=10. Value should be greater than 1.

• use_monotonic_fit – If True, eval scores in the eval dictionary are fitted to a monotonically increasing function. This is useful if you see the eval dict scores for some layers are not monotonically increasing. By default, this option is set to False.

• saved_eval_scores_dict – Path to the eval_scores dictionary pickle file that was saved in a previous run. This is useful to speed-up experiments when trying different target compression-ratios for example. aimet will save eval_scores dictionary pickle file automatically in a ./data directory relative to the current path. num_comp_ratio_candidates parameter will be ignored when this option is used.

Configuration Definitions

class aimet_torch.common.defs.CostMetric(value)

Enumeration of metrics to measure cost of a model/layer

mac = 1

Cost modeled for compute requirements

Type:

MAC

memory = 2

Cost modeled for space requirements

Type:

Memory

class aimet_torch.common.defs.CompressionScheme(value)

Enumeration of compression schemes supported in aimet

channel_pruning = 3

Channel Pruning

spatial_svd = 2

Spatial SVD

weight_svd = 1

Weight SVD

Channel Pruning Configuration

class aimet_torch.defs.ChannelPruningParameters(data_loader, num_reconstruction_samples, allow_custom_downsample_ops, mode, params, multiplicity=1)

Configuration parameters for channel pruning compression

class AutoModeParams(greedy_select_params, modules_to_ignore=None)

Configuration parameters for auto-mode compression

Parameters:

• greedy_select_params (GreedySelectionParameters) – Params for greedy comp-ratio selection algorithm

• modules_to_ignore (Optional[List[Module]]) – List of modules to ignore (None indicates nothing to ignore)

class ManualModeParams(list_of_module_comp_ratio_pairs)

Configuration parameters for manual-mode channel pruning compression

Parameters:

list_of_module_comp_ratio_pairs (List[ModuleCompRatioPair]) – List of (module, comp-ratio) pairs

class Mode(value)

Mode enumeration

auto = 2

AIMET computes optimal comp-ratio per layer

Type:

Auto mode

manual = 1

User specifies comp-ratio per layer

Type:

Manual mode