Guide for local ONNX model conversion to run on Copilot+ PCs?

Per the blog article

Fara-7B is available on (opens in new tab)Microsoft Foundry (opens in new tab)and (opens in new tab)Hugging Face(opens in new tab). We are also releasing the implementation of Fara-7B in Magentic-UI, so that users can try it in a contained environment through the inference code provided. Additionally, users can download the model for Copilot+ PCs powered by Windows 11 from the AI Toolkit in VSCode and run it all on-device, taking advantage of NPU hardware acceleration.

I've tried going through AI Toolkit and I see no way to download the model for Copilot+ PCs, and attempts at doing manual model conversion for NPUs using QNNExecutionProvider from AI Toolkit all fail.

Am I missing something? I'd love to see a step-by-step guide, since I don't see the model listed at all in the model searching capabilities in AI Toolkit either.

I really would love to test this model out locally and take advantage of the NPU.

Start with the Fara-7B inference code and the AI Toolkit model-search and QNNExecutionProvider paths mentioned in the issue. Verify how the model can be downloaded or converted for Copilot+ PCs, then document the required steps and a successful local run using NPU acceleration.