v2.2

Updated Torch and related dependency version. We are now at pytorch and lightning >= 2.0. Also support CUDA 11.7, 11.8, and 12.1.

Note

Export script is currently broken on torch 2.0 and 2.0.1. This is because onnx export doesn't work for multihead attention in this version. I've included the nightly torch release for 2.1 to make the export work. Alternatively users can downgrade to 1.13. Will create a new sub-release when torch 2.1 releases officially.

V2.1.1

Final release before updating torch versions.

V2.1

Export
To enable SPANet to be used outside of the python environment, we added an onnx export utility to allow for easy export to the onnx format which has run-times available for many languages. Please refer to this section of the documentation:

https://github.com/Alexanders101/SPANet#exporting

V2.0

1. New configuration file format with more options on inputs and event topology.
2. Allow for several different inputs, including global inputs for additional context.
3. New Regression and Classification output heads for performing per-event or per-particle predictions.
4. Gated transformers and linear layers for more robust networks. Less hyperparameter optimization.

v1.0

Original Release of code for reproducibility.