Support Qwen3.6-35B-A3B on Ryzen AI HX 370 / XDNA2

Hi FastFlowLM team,

First of all, thank you for your work on FastFlowLM. I’m very excited about the idea of running capable local LLMs efficiently on AMD Ryzen AI NPUs.

I have a system with:

• AMD Ryzen AI 9 HX 370
• 64 GB RAM
• XDNA2 / Ryzen AI NPU

I would really love to try Qwen3.6-35B-A3B on this hardware through FastFlowLM.

This model looks especially interesting because it is a 35B total / 3B active MoE model, so it seems like it could be a very good fit for efficient local inference if FastFlowLM can support it properly.

Would it be possible to add support for Qwen3.6-35B-A3B?

In particular, it would be amazing to have:

• official FastFlowLM support for Qwen3.6-35B-A3B
• a matching converter config for FLM Q4NX
• compatible XDNA2 / NPU2 xclbins for Ryzen AI HX 370 / Strix Point
• instructions for converting or running this model locally

I understand that this may require more than just converting the weights, especially if new xclbins or architecture-specific runtime support are needed. But I would be very happy to test it on my HX 370 + 64 GB RAM system and provide logs, feedback, benchmark results, or try experimental builds if that helps.

Qwen3.6-35B-A3B seems like one of the most interesting larger local models right now, and FastFlowLM support for it would be very exciting for AMD AI PC users.

Thanks again for the great project!

No files, tests, or entry points are named. Begin by reviewing existing Qwen support, converter configurations, and XDNA2/NPU2 xclbin handling; done would require a working Qwen3.6-35B-A3B path on Ryzen AI HX 370, with conversion or run instructions and validation on the reported hardware.