Examples

AIMET end-to-end examples are Jupyter Notebooks that are intended to:

• Familiarize you with the AIMET APIs,

• Demonstrate how to apply AIMET to a pre-trained model from PyTorch, TensorFlow and ONNX frameworks,

• Teach you how to use AIMET quantization and compression techniques.

For a discussion of quantization and compression techniques, see Optimization User Guide.

For the API reference, see API reference

Browse the notebooks

The following tables provide links to viewable HTML versions of the jupyter notebooks for AIMET quantization and compression features. Instructions after the tables describe how to run the notebooks.

Model Quantization Examples

Features	PyTorch	TensorFlow	ONNX
Quantization simulation (QuantSim)	Link	Link	Link
Quantization-aware training (QAT)	Link	Link	Not implemented.
Cross-layer equalization (CLE)	Link	Link	Link
Adaptive rounding (AdaRound)	Link	Link	Link
Automatic quantization (AutoQuant)	Link	Link	Not implemented.
Automatic mixed precision (AMP)	Link	Link	Link
BatchNorm re-estimation	Link	Link	Not implemented.
Quant analyzer	Link	Link	Not implemented.

Model Compression Examples

Features	PyTorch
Channel Pruning	Link
Spatial SVD	Link
Spatial SVD + Channel Pruning	Link

Running the notebooks

To run the notebooks, follow the instructions below.

1. Run the notebook server

1. Install the Jupyter metapackage using the following command. (Prepend the command with sudo -H if necessary to grant admin privilege.)

   python3 -m pip install jupyter
   

2. Start the notebook server as follows:

   jupyter notebook --ip=* --no-browser &
   

   The command generates and displays a URL in the terminal.

3. Copy and paste the URL into your browser.

4. Install AIMET and its dependencies using the instructions in AIMET installation.

2. Download the example notebooks and related code

Set up your workspace using the following steps:

1. Set a workspace path:

   WORKSPACE="<absolute_path_to_workspace>"
   

2. Create and move to the workspace:

   mkdir $WORKSPACE && cd $WORKSPACE
   

3. Identify the release tag (<release_tag>) of the AIMET package that you’re working with at: https://github.com/quic/aimet/releases.

4. Clone the repository:

   git clone https://github.com/quic/aimet.git --branch <release_tag>
   

5. Update the path environment variable:

   export PYTHONPATH=$PYTHONPATH:${WORKSPACE}/aimet
   

6. The dataloader, evaluator, and trainer used in the examples are for the ImageNet dataset. Download the ImageNet dataset from: https://www.image-net.org/download.php

3. Run the notebooks

1. Navigate to one of the following paths in your local repository directory and launch your chosen jupyter notebook (.ipynb extension):

Model quantization notebooks

• Examples/torch/quantization/

• Examples/tensorflow/quantization/keras/

• Examples/onnx/quantization/

Model compression notebooks

• Examples/torch/compression/

2. Follow the instructions in the notebook to execute the code.