Ollama released v0.34.1 with fixes targeting MLX runner memory management and token handling1. The update has mlxrunner check system free memory and wait for evicted runners before loading the next MLX model, while evicting prefix cache snapshots from the active conversation. The token repeat limit rises to 100, with errors now returned instead of incomplete results. Additional changes include keeping the gemma3n projector off the CPU, refreshing the Apps layout, and bundling MLX and llama.cpp updates.
Ollama v0.34.1 Patches MLX Memory, Raises Token Repeat Limit
Ollama v0.34.1 addresses MLX runner memory management, raises the token repeat limit to 100, and bundles updates to MLX and llama.cpp.