News for hackers. Stuff that matters.
Re:PyTorch at the Edge: Deploy 964 TIMM Models on Android wi... by skrunch on Wednesday February 15, 2023 @02:46AM (#34801154)
> PyTorch offers a way to export to ONNX but you will encounter various errors. [1]

I mean sure, there are limitations, but this is greatly exaggerating their impact in my experience. I'd be curious to hear from anyone where these have been serious blockers, I've been exporting PyTorch models to ONNX (for CV applications) for the last couple of years without any major issues (and any issues that did pop up were resolved in a matter of hours).

Re:PyTorch at the Edge: Deploy 964 TIMM Models on Android wi... by davidatbu on Wednesday February 15, 2023 @02:28PM (#34809281)
So I tried converting an ASR model[0] to ONNX about a year or two back. It was really painful. The pain could largely be ascribed to:

(1) code that is very dynamic, making it hard for Pytorch to convert the modules to TorchScript (which it does before converting them to ONNX)

(2) ops that were simply not available in ONNX. Especially, torch.fft, also some others.

[0] https://github.com/burchim/EfficientConformer