The latest release of ml-explore/mlx-swift is 0.32.3. It is a set of APIs that extends the machine learning array framework MLX for Apple silicon (Apple's own chips) into the Swift language, enabling GPU-accelerated model computation on Apple devices. The repository description is a single sentence — “Swift APIs for MLX” — and the recorded primary language is C++.
The point is to map MLX's array and neural network APIs onto Swift. If MLX is the framework that handles array and neural network computation, mlx-swift exposes that API as Swift types, so GPU-accelerated model computation can be called directly from Swift code. The README states that implementations for LLMs (language models) and VLMs (models that work with images and language together) are provided in the separate repository mlx-swift-lm.
The license is MIT. The creation date is recorded as December 12, 2023, and the star count as 2,066. Builds work with Xcode or SwiftPM, and the CMake install also provides a native Linux build option, so it is not a form limited to builds that assume only Apple silicon.
That said, what this 0.32.3 release changed is not documented in the given source record. Neither the changes nor the release date are presented, so the only thing confirmed here is the version number. In the scoring items, only the momentum score remains, at 2.1 points, while the freshness, technology and quality items are empty, and no performance measurements or run results are given. Whether it merely summarizes improvements on the MLX side or whether the Swift API surface changed can only be confirmed by checking the release notes directly.