Introducing nvFuser, a deep learning compiler for PyTorch (PyTorch blog)

Provenance for nvFuser's origin as MosaicML's fusion compiler: the post introduces nvFuser as 'a Deep Learning Compiler for NVIDIA GPUs that automatically just-in-time compiles fast and flexible kernels', notes it 'was specifically designed to meet the unique requirements of the PyTorch community', and states it targets 'Volta and later CUDA accelerators' while 'supporting diverse network architectures and programs with dynamic inputs of varying shapes and strides'. The documented flow is analyze, plan parallelization, apply, runtime-compile the generated GPU functions, execute the CUDA kernels. Also carries the honest coverage caveat: 'nvFuser does not yet support all PyTorch operations'. Historical context for the record's first_release_year and for why the toolchain binds so tightly to PyTorch versions.