A domain-specific language (DSL) is designed for a particular class of problems. Triton embeds its GPU-kernel DSL in Python, pairing the language with a compiler that turns decorated functions into GPU code. section 1.1.1
Decorating a Python function with @triton.jit lets developers express a kernel in Python and compile it on its first call. This keeps the workflow in Python while targeting GPU hardware; the matching runtime must still be installed. section 1.1.1
Triton occupies a middle ground between high-level frameworks and lower-level CUDA: it adds kernel-level control without requiring the full complexity of CUDA. section 1.5











