Ollama 0.34.1 Stabilizes MLX Backend and GGUF Model Creation

16 September 2026 1 min read

#apple #apple-silicon #gguf #inference-stability #memory-management #mlx #model-optimization #ollama #quantization #release

Ollama's latest stable release brings significant improvements to its MLX backend for Apple Silicon devices, including better memory management that should reduce crashes and improve throughput on Mac hardware. The release also graduates GGUF model creation from experimental status and refines token repetition detection to require 100 consecutive repeated tokens before triggering guards, reducing false positives in production scenarios.

These changes directly benefit local LLM operators: improved Apple Silicon support addresses a major deployment platform for individuals and small teams, while GGUF tooling maturity removes friction from model optimization workflows. The repeat token improvements matter for inference stability, especially with longer generation sequences or edge cases that can cause models to loop unexpectedly.

Read the full article on Ollama release.

Source: Ollama release · Relevance: 9/10