skills.rest
Just-it
Just-itCOMMUNITY
Explore all skills by this author →
GitHub

kernel-generator

Convert KernelBench operator descriptions into Triton Ascend kernels with a ModelNew class.

33|51|Updated Mar 12, 2026
Search Tags:#automation#iteration#code-generation#kernelbench#kernel-generation#triton-ascend
One-click install
npx skills add https://github.com/Just-it/AscendOpGenAgent --skill kernel-generator
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: kernel-generator
Source: https://github.com/Just-it/AscendOpGenAgent/tree/main/skills/kernel-generator
Command: npx skills add https://github.com/Just-it/AscendOpGenAgent --skill kernel-generator

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill automates the generation of high-performance Triton Ascend kernels from KernelBench style operator descriptions, reducing manual kernel development time and ensuring consistency with KernelBench specifications.

Core Features & Use Cases

  • Automated kernel generation: Produces a complete ModelNew-based Triton Ascend kernel implementation from a KernelBench task description.
  • Iterative improvement: Supports error feedback loop using verifier_error and conductor_suggestion to refine the kernel.
  • Knowledge integration: Loads hardware references and Triton Ascend fundamentals to guide optimization for ascend backends.
  • Use Case: Given a KernelBench op_desc, generate a full kernel plus a ModelNew class that mirrors the original model's forward signature.

Quick Start

Run the kernel-generator skill with a KernelBench task description to generate a complete ModelNew kernel.

Frequently Asked Questions about kernel-generator

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate Triton Ascend kernels from operator descriptions?▼

Triton Ascend kernel generation requires inputs like op_name, task_desc, and arch. It outputs a self-contained Python file with the kernel implementation and a ModelNew class mirroring the original forward signature.

Can I refine generated Triton kernels automatically using error feedback?▼

You can refine generated Triton kernels by supplying verifier_error and conductor_suggestion inputs. This error-driven loop automatically updates the kernel implementation to resolve identified issues.

What inputs are required to automate Triton Ascend kernel generation?▼

Kernel generation requires op_name, task_desc, and arch, while strictly enforcing backend=ascend and framework=torch. These inputs ensure the output aligns with KernelBench specifications.

Does Triton Ascend kernel generation work with frameworks other than PyTorch?▼

No, Triton Ascend kernel generation strictly enforces framework=torch. The output ModelNew class is specifically designed to integrate with PyTorch model architectures.

Why does my generated kernel include a ModelNew class?▼

The ModelNew class is included to mirror the original model's forward signature, ensuring seamless integration into existing PyTorch pipelines. This structure is required for KernelBench style implementations.

Related Skills

Explore More Software Engineering Skills→
mindspore-aimindspore-ai
258

kernel-generator

Generate operator kernels across multiple DSLs and backends.

Official
Advanced
xchang1121xchang1121
6

triton-ascend-example-layernorm

Implement a two-stage LayerNorm reduction kernel with Triton Ascend.

Community
Intermediate
mindspore-aimindspore-ai
258

triton-ascend-example-matmul

Demonstrate Triton Ascend matmul with 2D tiling and K-loop accumulation.

Official
Advanced
xchang1121xchang1121
6

kernel-workflow

Automate AI kernel generation, verification, and optimization across Triton, CUDA C, C++, and TileLang.

Community
Intermediate