If I had an eGPU right now, I'd 100% be using Qwen