to_onnx: another export to investigate

to_onnx implements another exporter to ONNX. It does not support all the cases torch.onnx.export(). It fails rather trying different options to recover. It calls torch.export.export() but does not alter the graph (no rewriting, no decomposition) before converting this graph to onnx. It is used to investigate export issues raised by torch.export.export().

Simple Case

• Export a linear regression

Dynamic Shapes

Dynamic shapes should be utilized to create a model capable of handling inputs with varying shapes while maintaining the same rank. Section torch.export.export: export to a Graph provides a couple of examples on how to define them, as their definition aligns with those used in torch.export.export().

• to_onnx and padding one dimension to a mulitple of a constant

Control Flow

• to_onnx and a model with a test

Custom Operators

• to_onnx and a custom operator registered with a function

• to_onnx and a custom operator inplace

Submodules

• to_onnx and submodules from LLMs

Optimization

• 101: Onnx Model Optimization based on Pattern Rewriting