This module contains high-quality helper classes and routines for building machine learning frameworks - but implemented in a framework-independent way. These routines help provide common currency types in the runtime, and aid implementation of glue code.
These libraries are framework independent, so they should not be very opinionated, and they should not depend on any specific framework modules.
This library provides common types for reasoning about Tensors with low-dependencies, suitable for use by both compiler and runtime clients. This is allowed to depend on LLVM Support, the Modular Support library, MLRT (Core, Driver, and GraphRT), the Mojo compiler, AsyncRT, and MLIR transforms. It is not allowed to depend on any specific framework's libraries (for example Faux, PyTorch etc).