Python-first compiler API
Provides an easy-to-use and customizable API for building machine learning compiler pipelines.
Universal deployment
Generates minimal deployable modules that can run on various hardware platforms including GPUs and edge devices.
Flexible compilation pipelines
Allows users to customize the compilation process to optimize models for specific hardware targets.
Support for multiple hardware backends
Includes backends such as CUDA, ROCm, Vulkan, OpenCL, and Metal to enable broad hardware compatibility.
Minimal runtimes
Enables efficient execution of machine learning workloads with minimal runtime overhead on both data center and edge hardware.