Thiago Crepaldi
Thiago Crepaldi
This should be fixed by [#109966](https://github.com/pytorch/pytorch/pull/109966), right?
@andreigh any update on this?
@fxmarty all shapes are static until you say otherwise using `input_names` and `dynamic_shapes` arguments. Do you have a full repro we could try? does it require a specific transformers version?...
Maybe depends on https://github.com/microsoft/onnxscript/issues/1462
* [Resolved] `export_params=None` is now properly errored out through `beartype` runtime checks, which will raise an exception if incorrect type is used * [Resolved] The problem with additional inputs was...
I am not sure. IIRC the exporter relies on onnx shape inference to validate and propagate the shape for the if. the exporter itself cannot make any guarantees there However,...
> [@BowenBao](https://github.com/BowenBao) any updates on this? Could you retest this, please?
FYI @wschin @xadupre
@svekars This is a great idea. https://github.com/pytorch/tutorials/pull/2541, for instance, is a tutorial that uses APIs only available on RC branches, so we cannot merge the PR before pytorch 2.1 goes...