Model save/load

[atlv24]

expose a way to save out a computation graph, accompanied by all constants parameters and model weights as relative-path hlo format files, xla-rs has utils for this and its the expected format for model/optimizer saving in XLA

[Ebanflo42]

xla-rs has utils for saving a converted XLA computation as HLO and also utils for saving literals as numpy, i think it would be dope to have these things combined in the same file, so the compute graph is serialized to HLO and eventually we have a Model object which contains weights that get serialized to a dictionary of numpy arrays.

[Ebanflo42]

Reading HLO into an unda context would be a neat trick. Might also be good to look into the ONNX format, I've never used this but apparently everyone is using it. Would boost our compatibility with existing APIs.

[Ebanflo42]

@BradenEverson or @atlv24 this should be implemented sooner rather than later if either of you have the bandwidth. There is already utilities for this in xla-rs/src/wrappers/mod.rs (of all places). If one of you could look into saving and loading SupervisedModel with these utilities I will clean up the organization of xla-rs.