tilelang-op-generate

To generate TileLang code, provide a design.md file detailing formulas, API mapping, shapes, tiling strategy, and verification cases. The skill parses this markdown and emits runnable Python TileLang operator implementation code with JIT decorators and test scaffolding.

Converting operator design specs involves parsing markdown documentation to extract formulas and tiling strategies, aligning specs with existing examples for idiomatic code, and emitting a complete example.py file with kernel logic and test harness ready for execution.

Yes, the design.md markdown structure should outline operator formulas, API mapping, shapes, tiling strategy, and verification cases. Parsing these specific details is required to synthesize and emit correct TileLang-Ascend operator implementations.

Yes, the generated TileLang operator code includes test scaffolding. The skill performs end-to-end generation to emit a complete Python file containing the kernel, JIT decorator, and a test harness ready for execution and verification.

Yes, you can use existing examples to align new TileLang operator generation. The skill features reference synthesis, which aligns extracted design specifications with existing examples to produce idiomatic TileLang-Ascend code.