Run Ornith, Qwen, Nemotron & DeepSeek optimized on AMD Strix Halo — unified OpenAI-compatible server with tuned ROCmFP4 quants, validated 262K context, vision support & hot-swap model zoo (iGPU + NPU)
