Backends
We want to leverage hardware acceleration on as many platforms as possible to optimize machine learning workloads. While the current support is limited to macOS with Metal, the framework is actively evolving to support a diverse range of computational devices.
Current Support
Below is a detailed table of supported backends and their current status:
| Backend | Description | Current Status |
|---|---|---|
| CPU | Central Processing Unit, available on all platforms. | ✅ Supported |
| Metal | macOS-specific GPU acceleration using Metal API. | ✅ Supported |
| CUDA | NVIDIA GPU acceleration using the CUDA API. | 🚧 Planned |
| OpenCL | Cross-platform GPU acceleration using OpenCL. | 🚧 Planned |
| OpenCL (CUDA) | Specialized OpenCL for NVIDIA GPUs. | 🚧 Planned |
| Vulkan | Cross-platform GPU support using Vulkan. | 🚧 Planned |
| DirectX12 | Windows GPU acceleration via DirectX 12. | 🚧 Planned |
| SYCL | Intel’s SYCL API for heterogeneous computing. | 🚧 Planned |
| TPUs | Google’s Tensor Processing Units for ML tasks. | 🚧 Planned |
| WebGPU | Lightweight GPU support for web-based environments. | 🚧 Planned |
On This Page