Is it possible to run models optimized for NPU with AMD processor?

[Maria Paula Coser Zamin (mpaulazamin)]

Hi!

After your help, I was able to download Hugging Face ONNX models. It worked for running on GPU.

Now, I'm trying to run models on NPU and Hybrid mode (NPU + GPU). I downloaded amd/Llama-3.2-3B-Instruct-awq-g128-int4-asym-bf16-onnx-ryzen-strix. However, I received the following error:

I pasted all the .dll libraries necessary to run models on NPU inside the model folder, but it did not work:

Is there a way I can run models on NPU that are optimized for AMD processor?

[Samuel Kemp (samuel100)]

Maria Paula Coser Zamin (@mpaulazamin) currently foundry local only supports QNN NPUs. We're working on automatic selection of execution providers that will open up more NPU providers, such as AMD. In the case of GPU - if you select the model-name-generic-gpu we'll use WebGPU that will select the GPU device.

[bmehta001]

Hi Maria Paula Coser Zamin (@mpaulazamin) , We now have models released for AMD NPUs in our catalog. Please open an issue if you have any questions.