Mozilla has released Llamafile 0.10, a significant update to their project that packages and runs large language models as a single cross-platform file. Key changes include a new build system, a hybrid TUI chat/server mode, a CLI modality for one-shot questions, integration of Whisper.cpp and Stable Diffusion as sub-modules, updated Llama.cpp, restored NVIDIA CUDA support, out-of-the-box Metal GPU support on macOS, a new --image argument for multimodal input, and improved BSD support.
Sort: