An open API service indexing awesome lists of open source software.

https://github.com/apple/ml-upscale

Export utility for unconstrained channel pruned models
https://github.com/apple/ml-upscale

deep-learning export machine-learning pruning

Last synced: 4 months ago
JSON representation

Export utility for unconstrained channel pruned models

Awesome Lists containing this project

README

          

# Unconstrained Channel Pruning · [Paper](https://openreview.net/forum?id=25fe54GXLo)

[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1vTZBWB3O2oj-g8oH5sj7j4CJ_xe6mv1p?usp=sharing)

**UPSCALE: Unconstrained Channel Pruning** @ [ICML 2023](https://openreview.net/forum?id=25fe54GXLo)

[Alvin Wan](https://alvinwan.com), [Hanxiang Hao](https://scholar.google.com/citations?user=IMn1m2sAAAAJ&hl=en&oi=ao), [Kaushik Patnaik](https://openreview.net/profile?id=~Kaushik_Patnaik1), [Yueyang Xu](https://github.com/inSam), [Omer Hadad](https://scholar.google.com/citations?user=cHZBEjQAAAAJ&hl=en), [David Güera](https://davidguera.com), [Zhile Ren](https://jrenzhile.com), [Qi Shan](https://scholar.google.com/citations?user=0FbnKXwAAAAJ&hl=en)

By removing constraints from existing pruners, we improve ImageNet accuracy for post-training pruned models by 2.1 points on average - benefiting DenseNet (+16.9), EfficientNetV2 (+7.9), and ResNet (+6.2). Furthermore, for these unconstrained pruned models, UPSCALE improves inference speeds by up to 2x over a baseline export.

## Quick Start

Install our package.

```bash
pip install apple-upscale
```

Mask and prune channels, using the default magnitude pruner.

```python
import torch, torchvision
from upscale import MaskingManager, PruningManager

x = torch.rand((1, 3, 224, 224), device='cuda')
model = torchvision.models.get_model('resnet18', pretrained=True).cuda() # get any pytorch model
MaskingManager(model).importance().mask()
PruningManager(model).compute([x]).prune()
```

## Customize Pruning

We provide a number of pruning heuristics out of the box:

- Magnitude ([L1](https://arxiv.org/abs/1608.08710) and [L2](https://arxiv.org/abs/1608.03665))
- [LAMP](https://arxiv.org/abs/2010.07611)
- [FPGM](https://arxiv.org/abs/1811.00250)
- [HRank](https://arxiv.org/abs/2002.10179)

You can pass the desired heuristic into the `UpscaleManager.mask` method call. You can also configure the pruning ratio in `UpscaleManager.mask`. A value of `0.25` means 25% of channels are set to zero.

```python
from upscale.importance import LAMP
MaskingManager(model).importance(LAMP()).mask(amount=0.25)
```

You can also zero out channels using any method you see fit.

```python
model.conv0.weight[:, 24] = 0
```

Then, run our export.

```python
PruningManager(model).compute([x]).prune()
```

## Advanced

You may want direct access to network segments to build a heavily-customized pruning algorithm.

```python
for segment in MaskingManager(model).segments():
# prune each segment in the network independently
for layer in segment.layers:
# layers in the segment
```

## Development

> **NOTE:** See [src/upscale/pruning/README.md](src/upscale/pruning/README.md) for more details on how the core export algorithm code is organized.

Clone and setup.

```bash
git clone git@github.com:apple/ml-upscale.git
cd upscale
pip install -e .
```

Run tests.

```
py.test src tests --doctest-modules
```

## Paper

Follow the development installation instructions to have the paper code under `paper/` available.

To run the baseline unconstrained export, pass `baseline=True` to `PruningManager.prune`.

```python
PruningManager(model).compute([x]).prune(baseline=True)
```

To reproduce the paper results, run

```bash
python paper/main.py resnet18
```

Plug in any model in the `torchvision.models` namespace.

```
usage: main.py [-h] [--side {input,output} [{input,output} ...]]
[--method {constrained,unconstrained} [{constrained,unconstrained} ...]]
[--amount AMOUNT [AMOUNT ...]] [--epochs EPOCHS]
[--heuristic {l1,l2,lamp,fpgm,hrank}] [--global] [--out OUT]
[--force] [--latency] [--clean]
model

positional arguments:
model model to prune

options:
-h, --help show this help message and exit
--side {input,output} [{input,output} ...]
prune which "side" -- producers, or consumers
--method {constrained,unconstrained} [{constrained,unconstrained} ...]
how to handle multiple branches
--amount AMOUNT [AMOUNT ...]
amounts to prune by. .6 means 60 percent pruned
--epochs EPOCHS number of epochs to train for
--heuristic {l1,l2,lamp,fpgm,hrank}
pruning heuristic
--global apply heuristic globally
--out OUT directory to write results.csv to
--force force latency rerun
--latency measure latency locally
--clean clean the dataframe
```

## Citation

If you find this useful for your research, please consider citing

```
@inproceedings{wan2023upscale,
title={UPSCALE: Unconstrained Channel Pruning},
author={Alvin Wan and Hanxiang Hao and Kaushik Patnaik and Yueyang Xu and Omer Hadad and David Guera and Zhile Ren and Qi Shan},
booktitle={ICML},
year={2023}
}
```