TIRx: An Open Compiler Stack for Evolving Frontier ML Kernels

Apache TVM introduces TIRx, an open-source hardware-native DSL and compiler for ML kernels targeting GPUs and specialized AI accelerators. Launch materials include a PyPI wheel and Python frontend with @T.jit/@T.prim_func authoring, a TIRx kernel library with GEMM and attention-style benchmarks on Blackwell GPUs, and an open course on modern GPU programming taught at Carnegie Mellon University.