Ollama v0.30.0 switches to llama.cpp, adds GGUF & MLX acceleration

Ollama has released v0.30.0 as a pre-release, marking a major architectural shift. Previously built on top of GGML, this version now directly integrates with llama.cpp, the C++ library behind many popular LLM backends. This change also introduces full compatibility with the GGUF file format, which has become the standard for quantized models in the llama.cpp ecosystem. For Apple Silicon users, the update adds MLX acceleration, Apple's machine learning framework optimized for M-series chips, promising faster inference on Macs. The release notes highlight that pre-release testing is underway, with the team specifically requesting feedback on performance improvements or regressions, errors or crashes not seen in earlier versions, and memory utilization changes.

However, the pre-release comes with known limitations. Two model types are explicitly unsupported: laguna-xs.2 and llama3.2-vision. Users relying on these models should hold off upgrading. The installation process remains straightforward: Mac/Linux users can run a curl command with OLLAMA_VERSION=0.30.0-rc22, while Windows users run a PowerShell script with the same version variable. The community has responded enthusiastically, with over 74 thumbs-up reactions and numerous other positive emoji responses on the release page. This update is significant because it aligns Ollama more closely with the broader open-source LLM ecosystem, potentially improving compatibility and performance for the thousands of developers who run models locally.