Siril 1.4.0beta3, GraXpert 3.1.0rc2, AI and Python

Hi everyone,

I’m using Siril 1.4.0beta3 with GraXpert 3.1.0rc2 on a PC equipped with a Core Ultra 7 265 and an NVIDIA RTX 5070.

When I run the GraXpert-AI.py script, I get the following error:
“WARNING: onnxruntime-directml is not built with CUDA 12.x support”

However, I installed the correct packages in Siril’s Python environment (torch 2.8.0+cu129 and onnxruntime-directml 1.22.0).

What am I doing wrong?

Many thanks in advance,
Best regards,
Philippe

You’re not doing anything wrong. There is a function call that is relevant to a different onnxruntime - it apparently produces a warning on your system but onnxruntime should still run fine and with acceleration. (If this isn’t the case let me know.)

At some point I’ll put the offending line in an if statement - I’m currently treating that as low priority compared with other 1.4.0 work, but if it actually is causing onnxruntime to be slow rather than just being a warning then I’ll see to it more urgently.

Thanks,

I suspected as much—the process completed successfully and produced a denoised image. Throughout the operation, the GPU fans were heavily active, which suggests that the AI was working.

Philippe