PMetal
A machine learning platform for Apple Silicon, written in Rust, covering local LLM inference, LoRA/QLoRA fine-tuning, quantization and serving through a desktop GUI, a terminal TUI and Metal/Neural Engine acceleration.
PMetal is a machine learning platform for Apple Silicon, written in Rust, spanning low-level Metal GPU kernels and Apple Neural Engine integration up to high-level training APIs, a terminal TUI and a full desktop GUI.
The desktop app, built with Tauri and Svelte, has pages for training, GRPO, distillation, pretraining, inference, quantization, dataset and model management and evaluation, letting a user download models from Hugging Face, configure LoRA training with live loss metrics, chat with a model, merge weights and quantize, all from the GUI. A companion terminal TUI covers the same ground across 20 tabs, including device and memory inspection, dataset scanning and tokenizing, and an OpenAI-compatible serving mode.
Training, inference, distillation and GRPO run in-process, while the remaining GUI pages drive the bundled command-line tool as a subprocess.