Apache TVM
An open-source machine learning compiler with a Python API that turns pretrained models into deployable modules for data center GPUs and edge devices.
Apache TVM is a machine learning compiler for developers optimizing models for deployment on their own hardware. It compiles model computations into modules that applications can embed, with targets spanning data center GPUs and edge devices. Code is licensed under Apache 2.0.
Its compiler stack combines Relax for graph-level representation with TensorIR for tensor programs. Python APIs let developers customize transformations and compilation pipelines rather than use a single fixed deployment path. Hardware support and the available optimization passes depend on the selected target and build configuration.
TVM also provides a runtime for executing compiled modules. This separates compilation work from the deployment environment and helps developers control the dependencies and runtime footprint of an application. It is a compiler and deployment toolkit, not a general model-training framework.
The official installation guide documents a CMake source build followed by installation of its Python package. Developers building for specialized targets need the corresponding compiler and hardware dependencies.
