Design UnifiedSketch DSL algorithm sketches for Triton Ascend kernels.
npx skills add https://github.com/Just-it/AscendOpGenAgent --skill kernel-designer
Please help me install this Agent Skill. Skill: kernel-designer Source: https://github.com/Just-it/AscendOpGenAgent/tree/main/skills/kernel-designer Command: npx skills add https://github.com/Just-it/AscendOpGenAgent --skill kernel-designer
Triton Ascend 算子算法草图设计 Skill — 根据任务描述设计高质量的算法草图（sketch），用于指导后续代码生成。支持首次设计和基于历史上下文的迭代优化。
Provide a task description and architecture to generate a UnifiedSketch design that guides subsequent kernel code generation.
High-intent search queries and answers about installing and using this skill.
To design algorithm sketches for Triton Ascend kernels, provide a task description and architecture to generate a UnifiedSketch DSL sketch that guides subsequent deterministic code generation. It applies to KernelBench style models within arch-specific constraints.
A UnifiedSketch DSL sketch is a high-level algorithm design used to guide kernel code generation for Triton Ascend. You need it when you want to produce deterministic kernel code based on fixed configurations like backend=ascend and framework=torch.
Yes, you can iteratively refine an existing Triton Ascend algorithm sketch by providing history_context and inspirations. The design process uses knowledge references to inform design patterns and optimize the sketch based on this historical context.
Yes, kernel sketch design operates with fixed configurations including backend=ascend and framework=torch. It specifically generates UnifiedSketch designs tailored for the Triton Ascend DSL within these architectural constraints.
The generated UnifiedSketch DSL sketch is consumed by a kernel-generator to produce deterministic Triton Ascend kernel code. You pass the sketch output directly to the kernel-generator to translate the algorithm design into functional code.
mindspore-aiDesign UnifiedSketch algorithm sketches for operator tasks across multiple DSLs.
mindspore-aiFuse multiple input slices into one output using Triton-Ascend kernels on Atlas hardware.
xchang1121Implement a two-stage LayerNorm reduction kernel with Triton Ascend.
mindspore-aiGenerate neural network operator sketches using UnifiedSketch DSL.
