Warpdrv - my open-source Llama.cpp launcher for daily-driving Qwen 35b + 27b on Strix Halo + RTX Pro.

I wanted to share an open-source app that I built for running LLMs locally on my setup. My setup Hardware FEVM FAEX1 (128GB) RTX Pro 5000 Blackwell (48GB), connected over OCuLink Aoostar AG02 2x2TB internal m.2 drives on raid-0 using mdadm. Software: Ubuntu 25.10, llama.cpp built from source for cuda + vulkan, rocm. How I use this app I generally run two models in parallel using different Llama backends simultaneously - Qwen3.6 27b UD-Q6-KXL or NVFP4 on CUDA, and Qwen3.6 35b A3B UD-