has this been tested on ORIN NX 8G too or only 16G version?

[monajalal]

[dataplayer12]

I only have a 16 GB Orin NX at hand so that's what I have tested on. Since the model doesn't use much VRAM it should absolutely work on the 8 GB NX too.

There is also a caveat. The HF --> ONNX conversion uses a lot of memory and even the 16 GB NX can't handle it. Please use any PC with enough RAM to export the ONNX and copy it to the 8GB Orin NX board. From there on, engine export and inference can run on that board without any memory limitations