Load LLM to CPU and offload to GPU

[Raboo]

I have a NAS with a Xeon Gold 6132 and 192 GB GRAM and a Geforce RTX 2080 TI 11GB VRAM GPU.

I want to use models that doesn't fit in my GPU memory. If I load larger models it fails because I assume it is trying to fit in the GPU VRAM.

I get errors like this.

Jan 14 19:19:48 INFO  BackendLoader starting modelID="mlabonne_gemma-3-27b-it-abliterated" backend="llama-cpp" model="mlabonne_gemma-3-27b-it-abliterated-Q4_K_M.gguf"
Jan 14 19:21:13 ERROR Failed to load model modelID="mlabonne_gemma-3-27b-it-abliterated" error=failed to load model with internal loader: could not load model: rpc error: code = Canceled desc =  backend="llama-cpp"
Jan 14 19:21:13 ERROR Stream ended with error error=failed to load model with internal loader: could not load model: rpc error: code = Canceled desc =

Is there a way to utilize both GPU and CPU and system memory instead of only VRAM for an LLM? Or am I forced to either or?
Also how do I tell LocalAI to load into CPU instead of GPU?
At the very least I thought it would be smart enough not to load LLMs into GPU that doesn't fit GPU memory.