floatingpoint64/torch-pruning

Towards Any Structural Pruning

Torch-Pruning

Towards Any Structural Pruning

Torch-Pruning (TP) is a general-purpose library for structural network pruning, which supports a large variaty of nerual networks like Vision Transformers, ResNet, DenseNet, RegNet, ResNext, FCN, DeepLab, VGG, etc. Please refer to tests/test_torchvision_models.py for more prunable models. Different from the torch.nn.utils.prune that only zeroizes parameters, Torch-Pruning, powered by a graph algorithm DepGraph, will physically removes parameters and sub-structures from your models.

Please refer to our preprint paper for more technical details: DepGraph: Towards Any Structural Pruning

Features:

• Structural (Channel) pruning for CNNs (e.g. ResNet, DenseNet, Deeplab) and Transformers (e.g. ViT)
• High-level pruners: MagnitudePruner, BNScalePruner, GroupPruner, RandomPruner, etc.
• Graph Tracing and dependency modeling.
• Supported modules: Conv, Linear, BatchNorm, LayerNorm, Transposed Conv, PReLU, Embedding, MultiheadAttention, nn.Parameters and customized modules.
• Supported operations: split, concatenation, skip connection, flatten, all element-wise ops, etc.
• Low-level pruning functions
• Benchmarks and tutorials

Plans:

• More high-level pruners like FisherPruner, SoftPruner, GeometricPruner, etc.
• Support more Transformers like Vision Transformers (:heavy_check_mark:), Swin Transformers, PoolFormers.
• More standard layers: GroupNorm, InstanceNorm, Shuffle Layers, etc.
• Examples for GNNs and RNNs.
• Pruning benchmarks for CIFAR, ImageNet and COCO.

Installation

pip install torch-pruning # v1.0.0

or

git clone https://github.com/VainF/Torch-Pruning.git

Quickstart

Here we provide a quick start for Torch-Pruning. More explained details can be found in tutorals

0. How it works

Dependency emerges in complicated network structures, which forces a group of parameters to be pruned simultaneouly. This works provides an automatical mechanism to group parameters with inter-depenedency, so that they can be correctly removed for acceleration. To be exact, Torch-Pruning will forward your model with a fake input and trace the network to establish a dependency graph, recording the dependency between layers. When you prune a single layer, Torch-pruning will also group those coupled layers by returning a Group. All pruning indices will be automatically transformed and aligned if there are operations like torch.split or torch.cat. We illustrate some dependencies in modern nerual networks as the following, where all highlighted channels and parameters will be removed together.

Dependency	Visualization	Example
Conv-Conv		AlexNet
Conv-FC (Global Pooling or Flatten)		ResNet, VGG
Skip Connection		ResNet
Concatenation		DenseNet, ASPP
Split		torch.chunk

1. A minimal example

import torch
from torchvision.models import resnet18
import torch_pruning as tp

model = resnet18(pretrained=True).eval()

# 1. build dependency graph for resnet18
DG = tp.DependencyGraph()
DG.build_dependency(model, example_inputs=torch.randn(1,3,224,224))

# 2. Specify the to-be-pruned channels. Here we prune those channels indexed by [2, 6, 9].
pruning_idxs = [2, 6, 9]
pruning_group = DG.get_pruning_group( model.conv1, tp.prune_conv_out_channels, idxs=pruning_idxs )

# 3. prune all grouped layer that is coupled with model.conv1 (included).
if DG.check_pruning_group(pruning_group): # avoid full pruning, i.e., channels=0.
    pruning_group.exec()

# 4. save & load the pruned model 
torch.save(model, 'model.pth') # save the model object
model_loaded = torch.load('model.pth') # no load_state_dict

This example shows the basic pipeline of pruning with DependencyGraph. Note that resnet.conv1 is coupled with several layers. Let's inspect the group and check how a pruning operation "triggers" other ones.

--------------------------------
          Pruning Group
--------------------------------
[0] [DEP] prune_out_channels on conv1 (Conv2d(3, 61, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)) => prune_out_channels on conv1 (Conv2d(3, 61, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)), #Pruned=3
[1] [DEP] prune_out_channels on conv1 (Conv2d(3, 61, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)) => prune_out_channels on bn1 (BatchNorm2d(61, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)), #Pruned=3
[2] [DEP] prune_out_channels on bn1 (BatchNorm2d(61, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)) => prune_out_channels on _ElementWiseOp(ReluBackward0), #Pruned=3
[3] [DEP] prune_out_channels on _ElementWiseOp(ReluBackward0) => prune_out_channels on _ElementWiseOp(MaxPool2DWithIndicesBackward0), #Pruned=3
[4] [DEP] prune_out_channels on _ElementWiseOp(MaxPool2DWithIndicesBackward0) => prune_out_channels on _ElementWiseOp(AddBackward0), #Pruned=3
[5] [DEP] prune_out_channels on _ElementWiseOp(MaxPool2DWithIndicesBackward0) => prune_in_channels on layer1.0.conv1 (Conv2d(61, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)), #Pruned=3
[6] [DEP] prune_out_channels on _ElementWiseOp(AddBackward0) => prune_out_channels on layer1.0.bn2 (BatchNorm2d(61, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)), #Pruned=3
[7] [DEP] prune_out_channels on _ElementWiseOp(AddBackward0) => prune_out_channels on _ElementWiseOp(ReluBackward0), #Pruned=3
[8] [DEP] prune_out_channels on _ElementWiseOp(ReluBackward0) => prune_out_channels on _ElementWiseOp(AddBackward0), #Pruned=3
[9] [DEP] prune_out_channels on _ElementWiseOp(ReluBackward0) => prune_in_channels on layer1.1.conv1 (Conv2d(61, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)), #Pruned=3
[10] [DEP] prune_out_channels on _ElementWiseOp(AddBackward0) => prune_out_channels on layer1.1.bn2 (BatchNorm2d(61, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)), #Pruned=3
[11] [DEP] prune_out_channels on _ElementWiseOp(AddBackward0) => prune_out_channels on _ElementWiseOp(ReluBackward0), #Pruned=3
[12] [DEP] prune_out_channels on _ElementWiseOp(ReluBackward0) => prune_in_channels on layer2.0.downsample.0 (Conv2d(61, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)), #Pruned=3
[13] [DEP] prune_out_channels on _ElementWiseOp(ReluBackward0) => prune_in_channels on layer2.0.conv1 (Conv2d(61, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)), #Pruned=3
[14] [DEP] prune_out_channels on layer1.1.bn2 (BatchNorm2d(61, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)) => prune_out_channels on layer1.1.conv2 (Conv2d(64, 61, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)), #Pruned=3
[15] [DEP] prune_out_channels on layer1.0.bn2 (BatchNorm2d(61, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)) => prune_out_channels on layer1.0.conv2 (Conv2d(64, 61, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)), #Pruned=3
--------------------------------

2. High-level Pruners

Based on DependencyGraph, we provide some high-level pruners in this repo for easy pruning. You can specify the channel sparsity to prune the whole model and fintune it using your own training code. Please refer to tests/test_pruner.py for more details. More examples can be found in benchmarks/main.py.

import torch
from torchvision.models import resnet18
import torch_pruning as tp

model = resnet18(pretrained=True)

# Importance criteria
example_inputs = torch.randn(1, 3, 224, 224)
imp = tp.importance.MagnitudeImportance(p=2)

ignored_layers = []
for m in model.modules():
    if isinstance(m, torch.nn.Linear) and m.out_features == 1000:
        ignored_layers.append(m) # DO NOT prune the final classifier!

iterative_steps = 5 # progressive pruning
pruner = tp.pruner.MagnitudePruner(
    model,
    example_inputs,
    importance=imp,
    iterative_steps=iterative_steps,
    ch_sparsity=0.5, # remove 50% channels, ResNet18 = {64, 128, 256, 512} => ResNet18_Half = {32, 64, 128, 256}
    ignored_layers=ignored_layers,
)

base_macs, base_nparams = tp.utils.count_ops_and_params(model, example_inputs)
for i in range(iterative_steps):
    pruner.step()
    macs, nparams = tp.utils.count_ops_and_params(model, example_inputs)
    # finetune your model here
    # finetune(model)
    # ...

3. Low-level pruning functions

Is is also possible to prune your model manually with low-level functions. But it would be quite effort-cunsuming since you need to carefully handle the dependency.

tp.prune_conv_out_channels( model.conv1, idxs=[2,6,9] )

# fix the broken dependencies manually
tp.prune_batchnorm_out_channels( model.bn1, idxs=[2,6,9] )
tp.prune_conv_in_channels( model.layer2[0].conv1, idxs=[2,6,9] )
...

The following pruning functions are available:

tp.prune_conv_out_channels,
tp.prune_conv_in_channels,
tp.prune_depthwise_conv_out_channels,
tp.prune_depthwise_conv_in_channels,
tp.prune_batchnorm_out_channels,
tp.prune_batchnorm_in_channels,
tp.prune_linear_out_channels,
tp.prune_linear_in_channels,
tp.prune_prelu_out_channels,
tp.prune_prelu_in_channels,
tp.prune_layernorm_out_channels,
tp.prune_layernorm_in_channels,
tp.prune_embedding_out_channels,
tp.prune_embedding_in_channels,
tp.prune_parameter_out_channels,
tp.prune_parameter_in_channels,
tp.prune_multihead_attention_out_channels,
tp.prune_multihead_attention_in_channels,

4. Customized Layers

Please refer to tests/test_customized_layer.py.

5. Benchmarks

Our results on {ResNet-56 / CIFAR-10 / 2.00x}

Method	Base (%)	Pruned (%)	$\Delta$ Acc (%)	Speed Up
NIPS [1]	-	-	-0.03	1.76x
Geometric [2]	93.59	93.26	-0.33	1.70x
Polar [3]	93.80	93.83	+0.03	1.88x
CP [4]	92.80	91.80	-1.00	2.00x
AMC [5]	92.80	91.90	-0.90	2.00x
HRank [6]	93.26	92.17	-0.09	2.00x
SFP [7]	93.59	93.36	+0.23	2.11x
ResRep [8]	93.71	93.71	+0.00	2.12x
Ours-L1	93.53	92.93	-0.60	2.12x
Ours-BN	93.53	93.29	-0.24	2.12x
Ours-Group	93.53	*93.91	+0.38	2.13x

Please refer to benchmarks for more details.

Citation

@article{fang2023depgraph,
  title={DepGraph: Towards Any Structural Pruning},
  author={Fang, Gongfan and Ma, Xinyin and Song, Mingli and Mi, Michael Bi and Wang, Xinchao},
  journal={arXiv preprint arXiv:2301.12900},
  year={2023}
}