v1.1.9: Bugfixing

Bugfixing for ViT, max_sparsity, etc.

v.1.1.8: Serialization for Pruned Models

1. Experimental Features: Save & load pruned models with tp.state_dict and tp.load_state_dict.
2. Bugfixing

Full Changelog: v1.1.7...v.1.1.8

v.1.1.7: Pruning via Taylor Expansion

• Add tp.importance.TaylorImportance
• Support PruningHistory
• Pruning & Post-training for Yolov8
• Bugfixing

v1.1.6

• An example for YOLOv8 pruning
• Improve Torchvision compatibility to 95.6% (81/85)
• Bugfix for torch.cat & torch.split

v1.1.5: Fixed a backward error

Update setup.py

v1.1.4

• Bugfix for reshape, view, flatten and nn.Parameter.
• Pre-trained models for CIFAR benchmark.

v1.1.3: BatchNorm / InstanceNorm / LayerNorm / GroupNorm

• Complete support for widely used normalization layers: Batch Normalization, Instance Normalization, Layer Normalization, and Group Normalization.
• Covers 90.6% of the models in Torchvision v0.13.1.
• Improved compatibility with PyTorch<=1.8.
• Bugfix in GroupNormPruner and Benchmark

v1.1.2: Automatic pruning of unwrapped nn.Parameters

• Automatic pruning of unwrapped nn.Parameter from non-standard layers.
• Bugfix (#135, #124, ...)

v1.1.1

1. Torchvision Compatibility: https://github.com/VainF/Torch-Pruning/tree/master/benchmarks/prunability
2. Reshape Support: Support for common reshape operations like .view, .reshape, and .flatten.
3. User-Friendly Interfaces: Easy-to-use interfaces for users of varying skill levels.
4. Robustness: Improve the robustness of Torch-Pruning
5. Yolov7: https://github.com/VainF/Torch-Pruning/tree/master/benchmarks/prunability/yolov7_detect_pruned.py
6. Bugfix

v1.0.0

Features:

• Channel pruning for CNNs (e.g. ResNet, DenseNet, Deeplab) and Transformers (e.g. ViT)
• High-level pruners: MagnitudePruner, BNScalePruner, GroupPruner, etc.
• Graph Tracing and dependency fixing.
• Supported modules: Conv, Linear, BatchNorm, LayerNorm, Transposed Conv, PReLU, Embedding, MultiheadAttention, nn.Parameters and customized modules.
• Supported operations: split, concatenation, skip connection, flatten, etc.
• Pruning strategies: Random, L1, L2, etc.
• Low-level pruning functions
• Benchmarks and tutorials