Hello, new guy here.
I’m running 4 Asrock AMD AI Pro R9700’s under vLLM. I really wanted to try Qwen3.8 Flash Next but it wasn’t running. I set Fable to work on it and after many tokens I got it running really well. If you’re trying something similar and want to see what I did you can check it out on my github: rickmellor/qwen3.8-flash-next-rdna4
I’m using this AWQ checkpoint on HF so I could do a partial offload to system RAM: leoncca/Qwen3.8-Flash-Next-AWQ-g32
I’m getting around 50 tok/s and its HumanEval, ARC, etc. scores are the best of any model I’ve tried. It’s highly usable and is my new daily driver.
Hopefully this is of use to someone.
./Rick