Fixed parsing of inputs for subgraphs

neuromorphs / NIRTorch

PyTorch helper module to translate to and from NIR

https://neuroir.org/docs

BSD 3-Clause "New" or "Revised" License

7 stars 3 forks source link

Fixed parsing of inputs for subgraphs #26

Closed Jegp closed 3 months ago

Jegp commented 4 months ago

This PR improves the parsing of subgraphs for Torch graphs by (1) explicitly parsing input nodes in the TorchGraph class and (2) moving some of the pruning of tensor nodes into the Graph tracer to avoid ambiguity.

I also rearrange the code slightly

Refactored out the GraphExecutor to a separate file with separate tests
Renamed Graph to TorchGraph and added tests for inputs
Added specialized functions for processing input nodes and edges
Updated Nix flakes

stevenabreu7 commented 3 months ago

Thanks a lot for the PR, Jens! This is much cleaner, I'm glad. I just fixed some unimportant test failures, should be good to merge now.

Jegp commented 3 months ago

Amazing, @stevenabreu7. Thank you for your feedback.