platform-limitations.md

Platform limitations

While ML.NET is cross-platform, there are some limitations for specific operating system and architecture combinations as outlined below.

Operating system and architecture	Training	Inference
Windows (x64)	Yes	Yes
Windows (x86)	Yes, with limitations. The following 64-bit-only components are not supported: TensorFlow-based components LightGBM training TorchSharp-based trainers	Yes, with limitations. The following 64-bit-only components are not supported: TensorFlow-based components TorchSharp-based models
Linux (x64)	Yes	Yes
macOS (Intel x64)	Yes	Yes
Windows, Linux or macOS (ARM64)	Yes, with limitations. The following are not supported using the native dependencies supplied by this repository: TensorFlow-based components OLS TimeSeries SSA TimeSeries SrCNN LightGBM TorchSharp-based trainers	Yes, with limitations. The following are not supported using the native dependencies supplied by this repository: TensorFlow-based components TimeSeries SSA TimeSeries SrCNN TorchSharp-based models
Blazor WebAssembly (browser)	Yes, with limitations. The following native-dependent components are not supported: Symbolic SGD TensorFlow-based components OLS TimeSeries SSA TimeSeries SrCNN ONNX LightGBM LDA Matrix Factorization TorchSharp-based trainers Note: You must currently set the EnableMLUnsupportedPlatformTargetCheck flag to false to use ML.NET in Blazor.	Yes, with limitations. The following native-dependent components are not supported: TensorFlow-based components TimeSeries SSA TimeSeries SrCNN ONNX LDA Matrix Factorization TorchSharp-based models

Symbolic SGD on macOS requires the Homebrew libomp package.

Failures for unsupported native components vary by package. Some are reported by an MSBuild platform check; others occur when the required native library cannot be loaded.

If you are blocked by any of these limitations or would like to see different behavior when hitting them, please let us know by filing an issue.