tvm

Apache TVM is an open machine learning compilation framework that focuses on Python-first development to allow for quick customization of compiler pipelines. It aims to provide universal deployment by converting models into minimum deployable modules. The current architecture utilizes a cross-level design featuring TensorIR for tensor-level representation and Relax for graph-level representation. This design allows for the joint optimization of computational graphs, tensor programs, and libraries. Additionally, TVM serves as foundational infrastructure for building domain-specific vertical compilers, including those for Large Language Models (LLMs).