madhusakth / model-compression-bq

Model-Compression-BQ

To create a virtual env and install Pytorch 1.9 cpu version

• python3 -m venv pytorch-venv
• source pytorch-venv/bin/activate
• pip3 install torch==1.9.0+cpu torchvision==0.10.0+cpu torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html

Run the gradient code train on a dummy CIFAR10 model -python3 model_gradients.py

This is the implementation of B&Q algorithm from Speech Recogniton Model Compression (https://ieeexplore.ieee.org/document/9053927). In the original paper, we tested our algorithm on 4 models: 1. Speech command recognition, 2. DeepSpeech, 3. DeepSpeech2, 4. VGG16.

In this repository, the automated code for B&Q will be shown for applying it as a standalone technique on multiple trained models.