Is there a separate Olive/ONNX optimization script?

[R-N]

Olive/ONNX optimization requires a lot of RAM and VRAM. My device can run the WebUI just fine with directml, but can't optimize checkpoints due to insufficient memory. So I want to use Google Colab just to optimize the checkpoint, but Colab prohibits WebUI and automatically terminates it. Is there a separate script for the optimization, perhaps? So I can use it in Colab.

By the way, I was assuming running the optimized checkpoint with Olive doesn't require significantly more RAM than normal checkpoint with just directml. Or does it?

Thank you

[lshqqytiger]

You have two options.

1. Download optimized model from huggingface instead of optimizing models yourself.
2. Try olive branch of vladmandic/automatic which requires less RAM.

[R-N]

I can't find olive-optimized version of the model I want to use.

I tried ripping the optimization script and running it on Colab. It didn't work as it needed DmlExecutionProvider and onnxruntime-directml. DirectML simply doesn't support linux because no directx.

Thanks for the alternative, I'll try it when I have the time.