Feature Request / Question: Support for Qwen3-Embedding-8B (.q4nx) on XDNA2

Hey team,

I'm looking to run local RAG workflows directly on the NPU (currently deploying on a Strix Halo APU) and would heavily prefer to use Qwen3-Embedding-8B to offload the embedding process.

I see that the core Qwen 3 and Qwen 3.5 architectures are already mapped and supported for text generation and vision models in the registry. However, I noticed there isn't an officially published .q4nx file for the Qwen3-Embedding-8B variant yet.

I have a couple of questions regarding the path forward:

Official Support: Are there plans to officially publish and maintain the .q4nx for this embedding model in the registry so it can be natively pulled via flm pull?

Custom Compilation: If I were to compile it myself, are there any known gotchas for this specific architecture? note its an embedding model

If there is a reference script or specific flm-build configuration you recommend for adapting the feature-extraction graph of Qwen 3, that would be incredibly helpful.

Thanks for the great work on the engine!

Johnny

Start with the Qwen 3 registry mappings and the flm-build configuration or feature-extraction graph referenced in the issue; verify how embedding models differ from text-generation models on XDNA2. Check whether a Qwen3-Embedding-8B .q4nx artifact can be built and pulled via flm pull, and document the required configuration or support decision.