Ollama v0.30.0-rc23: "directly support llama.cpp" & "compatibility with GGUF" (github.com)

Ollama has announced the pre-release of version 0.30.0-rc23, which introduces a significant architectural shift by directly supporting llama.cpp, moving away from its previous dependency on GGML. This version also offers compatibility with the new GGUF file format, enhancing the flexibility and usability of model deployments. Notably, this update includes the use of MLX to accelerate inference performance specifically on Apple Silicon, which could greatly benefit users leveraging this hardware for machine learning applications. The shift to direct support for llama.cpp is crucial for the AI/ML community as it streamlines model integration and enhances performance metrics. The feedback sought during this pre-release phase will help identify any performance improvements or issues related to inference speed, memory utilization, and overall stability. However, it's important to note that certain models, specifically laguna-xs.2 and llama3.2-vision, are not supported in this pre-release, indicating that users may need to wait for future updates to access full functionality.