leonardotredese / plinio

A Plug-and-play Lightweight tool for the Inference Optimization of Deep Neural networks

_{^{N.B., logo partially generated with Stable Diffusion.}}

---

PLiNIO is a Python package built on-top of the PyTorch ecosystem that provides a Plug-and-play Lightweight tool for the Inference Optimization of Deep Neural networks (DNNs).

PLiNIO allows to automatically optimize your DNN's architecture with no more than three additional lines of code to your original training loop.

PLiNIO exploits as main optimization engine Differentiable Neural Architecture Search (DNAS) algorithms which notoriusly balance flexibility and lightness.

At the current state, PLiNIO implements the following methods:

• PIT
• PIT-SuperNet
• Mixed Precision [N.B., this feature is experimental and currently under development]

You can consult the specific linked pages to gather more information on the usage and the capabilieties of such tools.

Installation

To install the latest release:

$ git clone https://github.com/eml-eda/plinio
$ cd plinio
$ python setup.py install

License

PLiNIO entire codebase is released under Apache License 2.0.