skmetal 0.2.0__tar.gz

skmetal-0.2.0/PKG-INFO

@@ -0,0 +1,292 @@
+Metadata-Version: 2.4
+Name: skmetal
+Version: 0.2.0
+Summary: GPU-accelerated scikit-learn via Apple Metal
+Author: Ainouche Abderahmane
+License-Expression: MIT
+Project-URL: Homepage, https://github.com/abderahmane-ai/skmetal
+Project-URL: Repository, https://github.com/abderahmane-ai/skmetal
+Project-URL: Bug Tracker, https://github.com/abderahmane-ai/skmetal/issues
+Keywords: scikit-learn,metal,gpu,apple-silicon,machine-learning
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: Operating System :: MacOS :: MacOS X
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: numpy>=1.24
+Requires-Dist: scikit-learn<2.0,>=1.5
+Requires-Dist: scipy>=1.10
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Requires-Dist: pytest-timeout>=2.3; extra == "dev"
+Requires-Dist: pytest-benchmark>=4.0; extra == "dev"
+Requires-Dist: matplotlib>=3.7; extra == "dev"
+Requires-Dist: pandas>=2.0; extra == "dev"
+
+# skmetal
+
+**Apple Silicon GPU acceleration for scikit-learn**
+
+[![Platform](https://img.shields.io/badge/platform-macOS%20Sonoma%2B-000000?logo=apple&style=flat-square)](https://github.com/abderahmane-ai/skmetal)
+[![Python](https://img.shields.io/badge/python-3.10%20|%203.11%20|%203.12-3776AB?logo=python&style=flat-square)](https://github.com/abderahmane-ai/skmetal)
+[![License](https://img.shields.io/badge/license-MIT-green?style=flat-square)](https://github.com/abderahmane-ai/skmetal)
+[![CI](https://github.com/abderahmane-ai/skmetal/actions/workflows/ci.yml/badge.svg)](https://github.com/abderahmane-ai/skmetal/actions/workflows/ci.yml)
+
+```python
+import skmetal
+from sklearn.linear_model import LinearRegression
+
+@skmetal.accelerate
+def model():
+    return LinearRegression()
+
+m = model()
+m.fit(X_train, y_train)
+m.predict(X_test)
+```
+
+---
+
+## Overview
+
+skmetal executes scikit-learn estimators on Apple Silicon GPUs via Metal Performance Shaders and custom Metal compute kernels. Decorate any function that returns an estimator with `@skmetal.accelerate` and `fit()`/`predict()` run on the GPU — no code changes required.
+
+Apple Silicon's unified memory architecture enables zero-copy data sharing: numpy arrays are passed directly to Metal via `bytesNoCopy`, eliminating data transfer overhead.
+
+---
+
+## Requirements
+
+- macOS 14+
+- Apple Silicon (M1-M5)
+- Python 3.10-3.12
+- Swift 6.1 (`xcode-select --install`)
+- scikit-learn >= 1.5
+
+---
+
+## Installation
+
+```bash
+pip install skmetal
+```
+
+> **Note**: macOS 14+ and Apple Silicon required. Xcode not needed for the pip package.
+
+### From source
+
+```bash
+git clone https://github.com/abderahmane-ai/skmetal.git
+cd skmetal
+bash build.sh
+pip install -e .
+```
+
+---
+
+## Usage
+
+### Decorator (recommended)
+
+```python
+import skmetal
+from sklearn.linear_model import LinearRegression
+
+@skmetal.accelerate
+def model():
+    return LinearRegression()
+
+m = model()
+m.fit(X, y)
+m.predict(X_test)
+```
+
+The decorator also works with pipelines:
+
+```python
+@skmetal.accelerate
+def pipeline():
+    return Pipeline([
+        ("scaler", StandardScaler()),
+        ("clf", LogisticRegression()),
+    ])
+
+pipe = pipeline()
+pipe.fit(X, y)
+pipe.predict(X_test)
+```
+
+### Function call
+
+```python
+model = skmetal.accelerate(LinearRegression())
+model.fit(X, y)
+```
+
+### Context manager
+
+```python
+with skmetal.accelerate_context():
+    model = LinearRegression()
+    model.fit(X, y)
+```
+
+---
+
+## Supported Estimators
+
+| Estimator | GPU Strategy | Speedup |
+|-----------|-------------|---------|
+| `LinearRegression` | Normal equations via MPS GEMM | **5.93x** |
+| `Ridge` | Fused centering + XTX + XTy (1 dispatch) | **1.16x** |
+| `LogisticRegression` | IRLS (3-5 Newton iterations, fused) | 0.91x |
+| `Lasso` | Coordinate descent + GPU residual updates | -- |
+| `ElasticNet` | Coordinate descent + GPU residual updates | -- |
+| `TruncatedSVD` | Randomized SVD, no centering (all BLAS-3) | **2.53x** |
+| `KMeans` | Single fused command buffer (all iterations on GPU) | 0.69x |
+| `DBSCAN` | GPU pairwise distance + per-point neighbor counting | -- |
+| `GaussianNB` | GPU mean/var per class | -- |
+| `StandardScaler` | Fused Welford (1 dispatch) | **8.27x** |
+| `MinMaxScaler` | Column min/max with threadgroup tree reduction | -- |
+| `RobustScaler` | GPU quantile approximation | -- |
+| `KNeighborsClassifier` | GPU pairwise distance + fused voting | -- |
+| `KNeighborsRegressor` | GPU pairwise distance + fused averaging | -- |
+| `NearestNeighbors` | GPU pairwise distance + index | -- |
+| `HistGradientBoostingRegressor` | C++ HGBT from sklearn (no custom GPU) | -- |
+| `HistGradientBoostingClassifier` | C++ HGBT from sklearn (no custom GPU) | -- |
+
+Estimators below 1.0x speedup are dispatch-limited at n <= 50K. Speedup improves to 2-5x at n >= 500K where compute dominates overhead.
+
+---
+
+## Architecture
+
+### Zero-copy pipeline
+
+```
+numpy array -> np.ctypes.data -> UnsafeMutableRawPointer -> MTLBuffer(bytesNoCopy:) -> GPU
+                   |                                                    |
+                   +--------- same physical memory ---------------------+
+```
+
+### Metal kernels (14 files)
+
+| Kernel file | Operations |
+|-------------|------------|
+| `ReductionKernels.metal` | `reduce_sum`, `reduce_mean_var` (Welford) |
+| `ArgminKernels.metal` | `argmin_rows` |
+| `KMeansKernels.metal` | assign, partial_sum, combine, normalize, batch fused |
+| `KNNKernels.metal` | tile top-k, merge, fused vote classify/regress |
+| `PairwiseDistKernels.metal` | `pairwise_distance_squared`, `pairwise_distance_direct` |
+| `DistanceKernels.metal` | `row_norm_sq`, `compute_mindists`, `distance_correct` |
+| `IrlsKernels.metal` | `irls_weight`, `scale_rows`, `compute_linear_irls`, `compute_error_scale`, `l2_reg_irls`, `multinomial_hessians` |
+| `CenterColumns.metal` | `column_means`, `center_columns` |
+| `ElementWiseKernels.metal` | sigmoid, subtract, add_scalar, axpy, norm_sq, transpose_f32, row_max, row_sum, softmax, negate |
+| `ExtraKernels.metal` | `soft_threshold`, `column_transform`, `scale_f32`, `sv_init`, `sv_hook`, `sv_shortcut` |
+| `StandardScalerKernels.metal` | `scaler_fit` (fused Welford) |
+| `GemmKernels.metal` | `gemm_simple` (fallback) |
+| `MinMaxKernels.metal` | `column_minmax` (threadgroup tree reduction) |
+| `TreeKernels.metal` | `tree_predict`, `tree_predict_all` |
+
+### Swift bridge (47 C-callable functions)
+
+All `skmetal_*` functions use `@_cdecl` for direct ctypes export. Every function accepts raw pointers.
+
+### Project structure
+
+```
+skmetal/
+  skmetal/
+    __init__.py
+    _bridge.py           ctypes -> Swift (47 functions)
+    _config.py           Config dataclass
+    _dispatch.py         estimator registry + wrapping
+    accelerate.py        @accelerate decorator + accelerate_context
+    estimators/
+      _base.py           BaseGPUEstimator abstract class
+      _registry.py       Estimator registry (17 estimators)
+      linear_model.py    LinearRegression, Ridge, LogisticRegression, Lasso, ElasticNet
+      cluster.py         KMeans, DBSCAN
+      decomposition.py   TruncatedSVD
+      ensemble.py        HistGradientBoostingRegressor, HistGradientBoostingClassifier
+      naive_bayes.py     GaussianNB
+      neighbors.py       KNeighborsClassifier, KNeighborsRegressor, NearestNeighbors
+      preprocessing.py   StandardScaler, MinMaxScaler, RobustScaler
+    utils.py
+  skmetal_bridge/        Swift + Metal
+    Sources/SkMetalBridge/
+      Bridge.swift       47 @_cdecl exports
+      MetalContext.swift
+      Kernels/*.metal    14 Metal kernel files
+  benchmarks/
+    run_compare.py       benchmark runner
+    benchmark_suite.py   full suite
+    baseline.json
+  tests/
+    test_correctness.py  18 estimator correctness tests
+    test_dispatch.py     7 dispatch logic tests
+    test_kernels.py      55 kernel unit tests
+    test_config.py       13 config API tests
+    test_accelerate.py   20 accelerate edge-case tests
+    test_stress.py       27 edge-case/stress tests
+    conftest.py          shared fixtures
+  build.sh
+  pyproject.toml
+  .github/workflows/
+    ci.yml               build + test + benchmark regression
+    release.yml          PyPI publish on tag
+  LICENSE                MIT
+```
+
+---
+
+## Tests
+
+```
+test_correctness   18/18 pass — estimator parametrizations + pipeline + device info
+test_dispatch       7/7  pass — registry, wrapping, pipeline, decorator
+test_kernels       55/55 pass — all 14 kernel files covered
+test_config        13/13 pass — config API, dtype rejection, threshold, reset
+test_accelerate    20/20 pass — decorator, context manager, edge cases
+test_stress        27/27 pass — contiguity, shape mismatch, extremes, CPU fallback
+```
+
+---
+
+## Benchmarks
+
+```bash
+python benchmarks/run_compare.py
+```
+
+---
+
+## Development
+
+```bash
+git clone https://github.com/abderahmane-ai/skmetal.git
+cd skmetal
+bash build.sh
+pip install -e ".[dev]"
+pytest tests/
+python benchmarks/run_compare.py
+```
+
+### Adding a new estimator
+
+1. Create `skmetal/estimators/my_model.py` with `MetalMyModel(BaseGPUEstimator)`
+2. Register in `_registry.py` (`GPU_ESTIMATORS` + `PIPELINE_PATTERNS`)
+3. Add module path in `_dispatch.py` (`module_map`)
+4. Write a Metal kernel if needed
+5. Add a parametrized test case in `test_correctness.py`
+
+---
+
+## License
+
+MIT License. See `LICENSE`.

skmetal-0.2.0/README.md

@@ -0,0 +1,262 @@
+# skmetal
+
+**Apple Silicon GPU acceleration for scikit-learn**
+
+[![Platform](https://img.shields.io/badge/platform-macOS%20Sonoma%2B-000000?logo=apple&style=flat-square)](https://github.com/abderahmane-ai/skmetal)
+[![Python](https://img.shields.io/badge/python-3.10%20|%203.11%20|%203.12-3776AB?logo=python&style=flat-square)](https://github.com/abderahmane-ai/skmetal)
+[![License](https://img.shields.io/badge/license-MIT-green?style=flat-square)](https://github.com/abderahmane-ai/skmetal)
+[![CI](https://github.com/abderahmane-ai/skmetal/actions/workflows/ci.yml/badge.svg)](https://github.com/abderahmane-ai/skmetal/actions/workflows/ci.yml)
+
+```python
+import skmetal
+from sklearn.linear_model import LinearRegression
+
+@skmetal.accelerate
+def model():
+    return LinearRegression()
+
+m = model()
+m.fit(X_train, y_train)
+m.predict(X_test)
+```
+
+---
+
+## Overview
+
+skmetal executes scikit-learn estimators on Apple Silicon GPUs via Metal Performance Shaders and custom Metal compute kernels. Decorate any function that returns an estimator with `@skmetal.accelerate` and `fit()`/`predict()` run on the GPU — no code changes required.
+
+Apple Silicon's unified memory architecture enables zero-copy data sharing: numpy arrays are passed directly to Metal via `bytesNoCopy`, eliminating data transfer overhead.
+
+---
+
+## Requirements
+
+- macOS 14+
+- Apple Silicon (M1-M5)
+- Python 3.10-3.12
+- Swift 6.1 (`xcode-select --install`)
+- scikit-learn >= 1.5
+
+---
+
+## Installation
+
+```bash
+pip install skmetal
+```
+
+> **Note**: macOS 14+ and Apple Silicon required. Xcode not needed for the pip package.
+
+### From source
+
+```bash
+git clone https://github.com/abderahmane-ai/skmetal.git
+cd skmetal
+bash build.sh
+pip install -e .
+```
+
+---
+
+## Usage
+
+### Decorator (recommended)
+
+```python
+import skmetal
+from sklearn.linear_model import LinearRegression
+
+@skmetal.accelerate
+def model():
+    return LinearRegression()
+
+m = model()
+m.fit(X, y)
+m.predict(X_test)
+```
+
+The decorator also works with pipelines:
+
+```python
+@skmetal.accelerate
+def pipeline():
+    return Pipeline([
+        ("scaler", StandardScaler()),
+        ("clf", LogisticRegression()),
+    ])
+
+pipe = pipeline()
+pipe.fit(X, y)
+pipe.predict(X_test)
+```
+
+### Function call
+
+```python
+model = skmetal.accelerate(LinearRegression())
+model.fit(X, y)
+```
+
+### Context manager
+
+```python
+with skmetal.accelerate_context():
+    model = LinearRegression()
+    model.fit(X, y)
+```
+
+---
+
+## Supported Estimators
+
+| Estimator | GPU Strategy | Speedup |
+|-----------|-------------|---------|
+| `LinearRegression` | Normal equations via MPS GEMM | **5.93x** |
+| `Ridge` | Fused centering + XTX + XTy (1 dispatch) | **1.16x** |
+| `LogisticRegression` | IRLS (3-5 Newton iterations, fused) | 0.91x |
+| `Lasso` | Coordinate descent + GPU residual updates | -- |
+| `ElasticNet` | Coordinate descent + GPU residual updates | -- |
+| `TruncatedSVD` | Randomized SVD, no centering (all BLAS-3) | **2.53x** |
+| `KMeans` | Single fused command buffer (all iterations on GPU) | 0.69x |
+| `DBSCAN` | GPU pairwise distance + per-point neighbor counting | -- |
+| `GaussianNB` | GPU mean/var per class | -- |
+| `StandardScaler` | Fused Welford (1 dispatch) | **8.27x** |
+| `MinMaxScaler` | Column min/max with threadgroup tree reduction | -- |
+| `RobustScaler` | GPU quantile approximation | -- |
+| `KNeighborsClassifier` | GPU pairwise distance + fused voting | -- |
+| `KNeighborsRegressor` | GPU pairwise distance + fused averaging | -- |
+| `NearestNeighbors` | GPU pairwise distance + index | -- |
+| `HistGradientBoostingRegressor` | C++ HGBT from sklearn (no custom GPU) | -- |
+| `HistGradientBoostingClassifier` | C++ HGBT from sklearn (no custom GPU) | -- |
+
+Estimators below 1.0x speedup are dispatch-limited at n <= 50K. Speedup improves to 2-5x at n >= 500K where compute dominates overhead.
+
+---
+
+## Architecture
+
+### Zero-copy pipeline
+
+```
+numpy array -> np.ctypes.data -> UnsafeMutableRawPointer -> MTLBuffer(bytesNoCopy:) -> GPU
+                   |                                                    |
+                   +--------- same physical memory ---------------------+
+```
+
+### Metal kernels (14 files)
+
+| Kernel file | Operations |
+|-------------|------------|
+| `ReductionKernels.metal` | `reduce_sum`, `reduce_mean_var` (Welford) |
+| `ArgminKernels.metal` | `argmin_rows` |
+| `KMeansKernels.metal` | assign, partial_sum, combine, normalize, batch fused |
+| `KNNKernels.metal` | tile top-k, merge, fused vote classify/regress |
+| `PairwiseDistKernels.metal` | `pairwise_distance_squared`, `pairwise_distance_direct` |
+| `DistanceKernels.metal` | `row_norm_sq`, `compute_mindists`, `distance_correct` |
+| `IrlsKernels.metal` | `irls_weight`, `scale_rows`, `compute_linear_irls`, `compute_error_scale`, `l2_reg_irls`, `multinomial_hessians` |
+| `CenterColumns.metal` | `column_means`, `center_columns` |
+| `ElementWiseKernels.metal` | sigmoid, subtract, add_scalar, axpy, norm_sq, transpose_f32, row_max, row_sum, softmax, negate |
+| `ExtraKernels.metal` | `soft_threshold`, `column_transform`, `scale_f32`, `sv_init`, `sv_hook`, `sv_shortcut` |
+| `StandardScalerKernels.metal` | `scaler_fit` (fused Welford) |
+| `GemmKernels.metal` | `gemm_simple` (fallback) |
+| `MinMaxKernels.metal` | `column_minmax` (threadgroup tree reduction) |
+| `TreeKernels.metal` | `tree_predict`, `tree_predict_all` |
+
+### Swift bridge (47 C-callable functions)
+
+All `skmetal_*` functions use `@_cdecl` for direct ctypes export. Every function accepts raw pointers.
+
+### Project structure
+
+```
+skmetal/
+  skmetal/
+    __init__.py
+    _bridge.py           ctypes -> Swift (47 functions)
+    _config.py           Config dataclass
+    _dispatch.py         estimator registry + wrapping
+    accelerate.py        @accelerate decorator + accelerate_context
+    estimators/
+      _base.py           BaseGPUEstimator abstract class
+      _registry.py       Estimator registry (17 estimators)
+      linear_model.py    LinearRegression, Ridge, LogisticRegression, Lasso, ElasticNet
+      cluster.py         KMeans, DBSCAN
+      decomposition.py   TruncatedSVD
+      ensemble.py        HistGradientBoostingRegressor, HistGradientBoostingClassifier
+      naive_bayes.py     GaussianNB
+      neighbors.py       KNeighborsClassifier, KNeighborsRegressor, NearestNeighbors
+      preprocessing.py   StandardScaler, MinMaxScaler, RobustScaler
+    utils.py
+  skmetal_bridge/        Swift + Metal
+    Sources/SkMetalBridge/
+      Bridge.swift       47 @_cdecl exports
+      MetalContext.swift
+      Kernels/*.metal    14 Metal kernel files
+  benchmarks/
+    run_compare.py       benchmark runner
+    benchmark_suite.py   full suite
+    baseline.json
+  tests/
+    test_correctness.py  18 estimator correctness tests
+    test_dispatch.py     7 dispatch logic tests
+    test_kernels.py      55 kernel unit tests
+    test_config.py       13 config API tests
+    test_accelerate.py   20 accelerate edge-case tests
+    test_stress.py       27 edge-case/stress tests
+    conftest.py          shared fixtures
+  build.sh
+  pyproject.toml
+  .github/workflows/
+    ci.yml               build + test + benchmark regression
+    release.yml          PyPI publish on tag
+  LICENSE                MIT
+```
+
+---
+
+## Tests
+
+```
+test_correctness   18/18 pass — estimator parametrizations + pipeline + device info
+test_dispatch       7/7  pass — registry, wrapping, pipeline, decorator
+test_kernels       55/55 pass — all 14 kernel files covered
+test_config        13/13 pass — config API, dtype rejection, threshold, reset
+test_accelerate    20/20 pass — decorator, context manager, edge cases
+test_stress        27/27 pass — contiguity, shape mismatch, extremes, CPU fallback
+```
+
+---
+
+## Benchmarks
+
+```bash
+python benchmarks/run_compare.py
+```
+
+---
+
+## Development
+
+```bash
+git clone https://github.com/abderahmane-ai/skmetal.git
+cd skmetal
+bash build.sh
+pip install -e ".[dev]"
+pytest tests/
+python benchmarks/run_compare.py
+```
+
+### Adding a new estimator
+
+1. Create `skmetal/estimators/my_model.py` with `MetalMyModel(BaseGPUEstimator)`
+2. Register in `_registry.py` (`GPU_ESTIMATORS` + `PIPELINE_PATTERNS`)
+3. Add module path in `_dispatch.py` (`module_map`)
+4. Write a Metal kernel if needed
+5. Add a parametrized test case in `test_correctness.py`
+
+---
+
+## License
+
+MIT License. See `LICENSE`.

skmetal-0.2.0/pyproject.toml

@@ -0,0 +1,63 @@
+[build-system]
+requires = ["setuptools>=68.0", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "skmetal"
+version = "0.2.0"
+description = "GPU-accelerated scikit-learn via Apple Metal"
+readme = "README.md"
+authors = [
+    {name = "Ainouche Abderahmane"},
+]
+requires-python = ">=3.10"
+license = "MIT"
+keywords = ["scikit-learn", "metal", "gpu", "apple-silicon", "machine-learning"]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Developers",
+    "Intended Audience :: Science/Research",
+    "Operating System :: MacOS :: MacOS X",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
+dependencies = [
+    "numpy>=1.24",
+    "scikit-learn>=1.5,<2.0",
+    "scipy>=1.10",
+]
+
+[project.optional-dependencies]
+dev = [
+    "pytest>=7.0",
+    "pytest-timeout>=2.3",
+    "pytest-benchmark>=4.0",
+    "matplotlib>=3.7",
+    "pandas>=2.0",
+]
+
+[project.urls]
+Homepage = "https://github.com/abderahmane-ai/skmetal"
+Repository = "https://github.com/abderahmane-ai/skmetal"
+"Bug Tracker" = "https://github.com/abderahmane-ai/skmetal/issues"
+
+[tool.setuptools.packages.find]
+where = ["."]
+namespaces = false
+
+[tool.setuptools.package-data]
+skmetal = ["*.dylib", "SkMetalBridge_SkMetalBridge.bundle/**/*"]
+
+[tool.ruff]
+target-version = "py310"
+line-length = 120
+extend-exclude = ["skmetal_bridge"]
+lint.select = ["E", "F", "W", "UP"]
+lint.ignore = ["E501", "E402"]
+
+[tool.pytest.ini_options]
+testpaths = ["../tests"]
+python_files = ["test_*.py"]
+timeout = 120

skmetal-0.2.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

skmetal-0.2.0/skmetal/__init__.py

@@ -0,0 +1,52 @@
+"""
+skmetal: Apple Silicon GPU acceleration for scikit-learn.
+
+Usage::
+
+    import skmetal
+    from sklearn.linear_model import LinearRegression
+
+    @skmetal.accelerate
+    def model():
+        return LinearRegression()
+
+    m = model()
+    m.fit(X, y)
+    m.predict(X_test)
+
+On non-Apple-Silicon machines skmetal imports cleanly and all estimators
+transparently fall back to scikit-learn CPU implementations.
+Check ``skmetal.METAL_AVAILABLE`` to detect GPU support at runtime.
+"""
+
+from ._about import __version__, __version_info__
+from ._config import get_config, set_device, set_threshold, set_dtype, set_verbose, set_thresholds, update_threshold
+from .accelerate import accelerate, accelerate_context
+from ._bridge import METAL_AVAILABLE
+
+if METAL_AVAILABLE:
+    from ._bridge import device_info
+else:
+    def device_info() -> dict:  # type: ignore[misc]
+        """Returns empty info when Metal is unavailable."""
+        raise RuntimeError(
+            "skmetal: device_info() requires Apple Silicon + macOS 14+. "
+            "Metal is not available on this device."
+        )
+
+__all__ = [
+    "__version__",
+    "__version_info__",
+    "METAL_AVAILABLE",
+    "accelerate",
+    "accelerate_context",
+    "device_info",
+    "get_config",
+    "set_device",
+    "set_threshold",
+    "set_dtype",
+    "set_verbose",
+    "set_thresholds",
+    "update_threshold",
+]
+

skmetal-0.2.0/skmetal/_about.py

@@ -0,0 +1,2 @@
+__version__ = "0.2.0"
+__version_info__ = (0, 2, 0)