nexport is a lightweight
Python 3.10+ package which empowers deep learning developers in exporting the trainable parameters of deep neural networks to human-readable and transrerable file types.
Table of contents
Install & use
- From terminal:
pip install nexport
- From python environment:
- Export weights and biases to human-readable file
- Ensure compatability with all popular neural network development software
This package is intended to be used in conjunction with inference-engine. As such,
nexport was developed by the
inference-engine developers to enable compatability between the two softwares.
nexport does this by exporting the weights and biases from networks compiled in
TensorFlow into standardized human-readable files. These files can be read by
inference-engine to instantiate the netwoks in Fortran 2018 for inference.
nexport was created and is currently maintained by Jordan Welsman. Parts of this project were based on prior work by Tan Nguyen.
nexport is developed and distributed under a Berkeley Laboratory modified
LICENSEfor more details.