OllamaFrom the builder
Faster Gemma 4 on MLX with multi-token prediction
Ollama 0.31 turns on multi-token prediction (MTP) for Gemma 4, where a small draft model proposes the next few tokens and the main model checks them in one pass; it is on by default and leaves output unchanged. On a coding-agent run of the Aider polyglot benchmark, Gemma 4 generated tokens nearly 90% faster on average on Apple Silicon. Ollama also contributed an MLX kernel for small verification batches that makes Gemma 4's largest matrix multiplications 2x to 2.5x faster on an M5 Max with nvfp4.
Code is especially predictable. It is full of closing brackets, repeated identifiers, and boilerplate, so the draft model’s proposals are accepted often.
Saved . A shortened excerpt from the official blog; the original publisher retains authorship.