This release migrates torch graph parsing and NIR interpreting to `torch.fx`. Specifically, we introduce new API functions `nir_to_torch` and `torch_to_nir` and deprecates the old `load` and `extract_nir_graph` functions.
The main advantages of torch.fx is
1. Simplicity, since their API is close to a raw graph (which aligns well with NIR) and
2. Tracing that is purely symbolic so we can avoid passing mock data into the parsing functions.
3. Additionally, this lets us leverage pattern matching and similar abstract graph operations in the future.
What's Changed
* Added license and citation by Jegp in https://github.com/neuromorphs/NIRTorch/pull/24
* Fixed parsing of inputs for subgraphs by Jegp in https://github.com/neuromorphs/NIRTorch/pull/26
* Ports the graph tracing to torch.fx by Jegp in https://github.com/neuromorphs/NIRTorch/pull/28
* Updated README and version numbering by Jegp in https://github.com/neuromorphs/NIRTorch/pull/29
**Full Changelog**: https://github.com/neuromorphs/NIRTorch/compare/v1.0...v2.0