Keras的模型压缩CLI工具
hard_compressor
用于keras的模型压缩CLI工具。
如何使用它
环境要求
Python 3.5, 3.6
Keras
安装方法
$ git clone ${this repository}
$ cd ./keras_compressor
$ pip install .
模型压缩
示例
keras-compressor.py model.h5 compressed.h5
带有精度参数
keras-compressor.py --error 0.001 model.h5 compressed.h5
帮助
$ keras-compressor.py --help
Using TensorFlow backend.
usage: keras-compressor.py [-h] [--error 0.1]
[--log-level {CRITICAL,ERROR,WARNING,INFO,DEBUG}]
model.h5 compressed.h5
compress keras model
positional arguments:
model.h5 target model, whose loss is specified by
`model.compile()`.
compressed.h5 compressed model path
optional arguments:
-h, --help show this help message and exit
--error 0.1 layer-wise acceptable error. If this value is larger,
compressed model will be less accurate and achieve
better compression rate. Default: 0.1
--log-level {CRITICAL,ERROR,WARNING,INFO,DEBUG}
log level. Default: INFO
压缩demo
在示例目录中,您将找到使用MNIST和CIFAR10数据集的类似VGG的模型的模型压缩。
$ cd ./keras_compressor/example/mnist/
$ python train.py
-> outputs non-compressed model `model_raw.h5`
$ python compress.py
-> outputs compressed model `model_compressed.h5` from `model_raw.h5`
$ python finetune.py
-> outputs finetuned and compressed model `model_finetuned.h5` from `model_compressed.h5`
$ python evaluate.py model_raw.h5
$ python evaluate.py model_compressed.h5
$ python evaluate.py model_finetuned.h5
-> output test accuracy and the number of model parameters