Instructions to use alastorid/WaifuRuntime-Metal with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use alastorid/WaifuRuntime-Metal with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download alastorid/WaifuRuntime-Metal --local-dir WaifuRuntime-Metal
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
WaifuFinder native Metal runtime
Compiled Metal kernels for the WaifuFinder native Swift runtimes. These are build
artifacts, not model weights โ model weights live in their own repositories, such as
alastorid/WD-SwinV2-Tagger-v3-Swift-MLX.
| File | Purpose |
|---|---|
mlx.metallib |
Official MLX 0.31.1 Metal kernels, matching the core bundled by mlx-swift 0.31.2. Extracted kernels only: no Python interpreter, wheel installation or Python runtime is required. |
Why this is hosted separately
The WaifuFinder build compiles a Swift/Metal worker and needs this precompiled kernel
library at build time. It is not a weight file and is never downloaded by end users at
runtime โ the released app bundles it inside Contents/Resources/Runtime/.
Hosting it on the Hub gives the build a stable, revision-pinned, checksum-verified source that does not depend on a GitHub release page.
Integrity
mlx.metallib is pinned by SHA-256:
928bdaf4d28683442b31bf11760f4dec600fe705e2a6367aaa3de0c5801d2066 mlx.metallib
tools/build-native-uncover.sh verifies this digest on every native build and refuses
to continue on mismatch.
Licence
The kernels are from MLX, MIT licensed. The full
terms travel in MLX-LICENSE.