[MrDrMcCoy]

My attempts to try ternary encodings from Unsloth with llama.cpp on ROCm failed miserably. Either ggml or ROCm simply can't run it at this time on gfx1201, and CPU isn't fast enough.