ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
A development skill for writing complete PyPTO operators from requirements, designs, and reference implementations. It also creates related tests and documentation.
ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
A PyPTO skill for analyzing and automatically tuning operators, including running test cases, checking numerical accuracy, collecting performance data, and producing a performance report.
ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
A performance-analysis skill for PyPTO operators that extracts important measurements from performance data files, rates performance, and identifies bottlenecks.
ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
A code-level performance-tuning skill for PyPTO operators, focused on how frontend code is written. It covers loops, tile-shape settings, and data operations.
ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
A PyPTO skill for improving the performance of one operator task by examining its instructions and operations. It covers instruction-level changes, internal processing stages, and unusual input shapes.
ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
A deep-performance tuning skill for PyPTO operators that uses a lane diagram to examine how work is scheduled and overlaps. It covers Stitch, tile-shape, operator fusion, and scheduling adjustments.
ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
A PyPTO debugging skill for comparing numerical results using saved files or checkpoint tensors. A checkpoint tensor is an intermediate result captured during execution.
ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
A troubleshooting skill for finding numerical-accuracy problems in PyPTO operators. The supplied description identifies its purpose but does not list its specific debugging steps.
ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
A connection layer that lets an existing Ascend C operator project be called from PyTorch through torch.ops.npu.xxx(). Ascend C is used to write kernels for Ascend NPU hardware.
ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
An optimization skill for Triton operators running on Ascend NPU hardware. Triton is a language for writing GPU- and accelerator-focused data-processing code.
ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
A code-generation guide for creating Triton Ascend kernels from operator task descriptions, then improving them from error feedback. Triton Ascend is a language and platform for writing accelerated computation kernels.
ascend-ai-coding/awesome-ascend-skills
Skill Claude CodeCodex
An algorithm-planning guide for designing sketches for Triton Ascend operators from task descriptions or earlier sketches. A sketch is a high-level plan used to guide later kernel implementation.