When running bigger styles that don't match into VRAM on macOS, Ollama will now split the model among GPU and CPU To maximise performance. Meta says that Llama 3 outperforms competing versions of its course on important benchmarks and that it’s superior through the board at duties like coding. https://finngbbnm.thelateblog.com/27174548/5-essential-elements-for-wizardlm-2