An in-depth exploration of JIT decorator patterns in Python machine learning libraries showcases three implementation strategies: AST-based, bytecode-based, and tracing-based JIT compilation. Each approach offers unique benefits for translating Python functions into optimized machine code, with detailed examples from JAX and Triton demonstrating real-world applications.
https://eli.thegreenplace.net/2025/decorator-jits-python-as-a-dsl/