No idea. MLX is its own thing, optimized for Apple Silicon. Ollama uses GGUFs.
https://ollama.com/library/gemma3:27b-it-qat says it's Q4_0. https://huggingface.co/mlx-community/gemma-3-27b-it-qat-4bit says it's 4bit. I think those are the same quantization?