triton-operator-precision-eval

Name: triton-operator-precision-eval
Availability: InStock
Author: Ascend

Benchmark Triton operator precision against PyTorch references and generate MERE/MARE reports.

31|1|Updated Mar 28, 2026

Search Tags:#verification #comparison #benchmark #precision #operator #triton

Ready to run

npx skills add https://github.com/Ascend/agent-skills --skill triton-operator-precision-eval

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: triton-operator-precision-eval
Source: https://github.com/Ascend/agent-skills/tree/main/skills/triton-operator-precision-eval
Command: npx skills add https://github.com/Ascend/agent-skills --skill triton-operator-precision-eval

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires triton, torch, torch_npu, pytest, numpy, and includes scripts (resource) components.

What problem does it solve?

Triton 算子精度评估。通过与 PyTorch 参考实现对比，自动计算误差指标并生成标准化精度报告，确保在进行性能优化前算子数值正确性。

Core Features & Use Cases

基于 Torch 参考实现（标杆）与多种 shape × dtype 组合的对比，生成一致的精度报告。
自动化输出：包括 MERE、MARE、最大相对误差等指标，以及可复现的精度报告。
使用场景：验证新 Triton 内核在生产管线中的数值一致性，保障与 PyTorch 参考实现的一致性。

Quick Start

Run the precision evaluation to compare Triton operator outputs against PyTorch references and generate the precision report.

Frequently Asked Questions about triton-operator-precision-eval

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

FAQPage Schema

How do I verify Triton operator precision against a PyTorch reference implementation?▼

You can verify Triton operator precision by benchmarking outputs against PyTorch reference implementations across diverse shapes, dtypes, and epsilon settings to produce a standardized accuracy report.

What metrics are used for Triton kernel accuracy verification?▼

Triton kernel accuracy verification uses MERE, MARE, and maximum relative error metrics to calculate error against PyTorch references and generate a reproducible precision report.

How do I automate precision evaluation for a fused Triton kernel?▼

Automating precision evaluation for a fused Triton kernel involves running an end-to-end workflow that compares kernel outputs with PyTorch references and automatically generates a report with error metrics.

Do I need pytest and numpy to benchmark Triton operators?▼

Yes, you need pytest and numpy, along with triton, torch, and torch_npu, to execute the benchmarking scripts that compare operator outputs and validate numerical consistency.

What is the best way to ensure numerical consistency before optimizing Triton operators?▼

The best way to ensure numerical consistency before performance optimization is to run automated precision verification that validates your Triton operator against a PyTorch reference implementation across multiple configurations.

Why does my Triton operator output mismatch the PyTorch reference across different shapes?▼

Mismatches across different shapes occur because precision varies by dtype and shape configurations; running a comprehensive benchmark evaluation identifies these specific numerical inconsistencies.