OllamaFrom the builder
Ollama is now powered by MLX on Apple Silicon in preview
Ollama released a preview that rebuilds its Apple silicon backend on MLX, Apple's machine learning framework, to use unified memory, and on M5-series chips it taps the new GPU Neural Accelerators for faster time to first token and generation. The preview also supports NVIDIA's NVFP4 format and upgrades caching with reuse across conversations and snapshots at key points in the prompt. Ollama says version 0.19 reaches 1851 tokens per second prefill and 134 tokens per second decode with int4 quantization, and it requires a Mac with more than 32GB of unified memory.
As more inference providers scale inference using NVFP4 format, this allows Ollama users to share the same results as they would in a production environment.
Saved . A shortened excerpt from the official blog; the original publisher retains authorship.