Engines · Metal
Apple Silicon.
Metal execution on Apple Silicon Macs. Language models run natively on the GPU from your own checkpoint, with no Python environment and no converted model copy.
Metal
The Apple Silicon backend.
What it is, in one view.
The checkpoint you already have is read as published; the compiler lowers it to native code and the engine runs it through the backend’s kernels on your hardware. Nothing is uploaded.
This backend runs through the same engine and compiler as every other — only the kernels and the device change. The platform overview has the shared path.
Execution path
Checkpoint in, native execution out.
The engine reads the published checkpoint and the compiler lowers it to native Metal compute for the Mac’s GPU; the run executes on the unified-memory device directly.
Features & install
What it does, and what it needs.
| Platform | macOS on Apple Silicon (arm64) |
|---|---|
| Backend | Metal, on the Apple Silicon GPU. |
| Chip generations | M1, M2, M3, M4 — see the Apple Silicon table. |
| Model classes | Language today. The language catalogue lists each family and its licence. |
| Features | Command-line generation, interactive chat and a local OpenAI-compatible HTTP server for your own tools. |
| Install conditions | Apple Silicon (M1 or newer), and macOS 13 or later. It is not tied to one macOS version: 13 is the minimum tested, and newer releases run. Activate each Mac once in your browser (veizik activate); generation then runs offline. Model weights are not included — you supply the public checkpoint. |