I can't find documentation how to use GPU so it's using a lot of CPU and very slow. This happens with all local models including llm-gguf plugin.