Mit Auto Tune Models

1. Voice Changer
2. Mit Auto Tune Models 2017

Our Services – All Makes & Models. We are one of the few remaining full-service automotive repair facilities in the area. Since 1981, THE AUTO SHOP has continually grown and improved to better serve our Clients.Our desire is to cultivate a continued relationship with you, your family, and your friends, meeting all your vehicles needs.

This repo contains the PyTorch implementation for paper AMC: AutoML for Model Compression and Acceleration on Mobile Devices.

Voice Changer

Reference

If you find the repo useful, please kindly cite our paper:

Other papers related to automated model design:

• HAQ: Hardware-Aware Automated Quantization with Mixed Precision (CVPR 2019)

• ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware (ICLR 2019)

Training AMC

Current code base supports the automated pruning of MobileNet on ImageNet. The pruning of MobileNet consists of 3 steps: 1. strategy search; 2. export the pruned weights; 3. fine-tune from pruned weights.

To conduct the full pruning procedure, follow the instructions below (results might vary a little from the paper due to different random seed):

1. Strategy Search

   To search the strategy on MobileNet ImageNet model, first get the pretrained MobileNet checkpoint on ImageNet by running:

   It will also download our 50% FLOPs compressed model. Then run the following script to search under 50% FLOPs constraint:

   Results may differ due to different random seed. The strategy we found and reported in the paper is:

2. Export the Pruned Weights

   After searching, we need to export the pruned weights by running:

   Also we need to modify MobileNet file to support the new pruned model (here it is already done in models/mobilenet.py)

3. Fine-tune from Pruned Weightsa

   After exporting, we need to fine-tune from the pruned weights. For example, we can fine-tune using cosine learning rate for 150 epochs by running:

AMC Compressed Model

We also provide the models and weights compressed by our AMC method. We provide compressed MobileNet-V1 and MobileNet-V2 in both PyTorch and TensorFlow format here.

Detailed statistics are as follows:

Dependencies

Current code base is tested under following environment:

1. Python 3.7.3
2. PyTorch 1.1.0
3. torchvision 0.2.1
4. NumPy 1.14.3
5. SciPy 1.1.0
6. scikit-learn 0.19.1
7. ImageNet dataset

Contact

To contact the authors:

Mit Auto Tune Models 2017

Ji Lin, jilin@mit.edu

Song Han, songhan@mit.edu