QTensor

QTensor is an extended repository for Quantization-Aware Training (QAT) with TensorFlow. Focusing on using different quantizers with the goal of benchmarking different ones.

Developing

To get started with QTensor, clone the repository and run:

1. Clone the repo

git clone https://github.com/username/qtensor.git
cd qtensor

2. Run the docker container

./docker/run.sh

3. Run examples

• Select a model under the models module, fo example mlp.
• Select one of the qconfigs associated with the selected model (the available ones are the keys of the qconfigs dictionary e.g. for mlp).
• Select a dataset to use to evaluate the model from the available datasets.

cd src/examples
./run.py --model mlp --qconfig qconfig --dataset mnist`

Note: Run ./run.py --help for extra options

Contributing

Working on the thesis article

To run and develop the document with latex, you can compile the file by running compile.sh.

If you are working with VSC

If you use visual studio code as your IDE you can customize it so ctrl+enter compiles the latex document.

1. add this task.json file to the .vscode dir in the root of the project.

{
    "version": "2.0.0",
    "tasks": [
        {
            "label": "compile latex",
            "type": "shell",
            "command": "docker/compile.sh",
            "presentation": {
                "echo": false,
                "reveal": "silent",
                "focus": false,
                "panel": "shared",
                "showReuseMessage": false,
                "clear": true,
            }
        }
    ]
}

2. Add this shortcut to your keybindings

[{
    "key": "ctrl+enter",
    "command": "workbench.action.tasks.runTask",
    "args": ["compile latex"]
  }
]

How to contribute

Please before pushing, install pre-commits and run them

pip install pre-commit==2.20.0
pre-commit install
pre-commit run --all-files --verbose --show-diff-on-failure