Native ROCm acceleration for Ollama on the AMD Radeon 780M iGPU (gfx1103) on Ubuntu. One-command setup that builds ollama-for-amd from source, applies three small ml/device.go patches, and installs Fedora 43's prebuilt gfx1103 Tensile kernels into the system rocBLAS.

llama.cpp fork — gfx1103 / Radeon 780M MMVQ nwarps decode fix (+22–36% tg). See GFX1103‑780M‑FIX.md