KernelGenBench Documentation#
A benchmark framework for evaluating LLM and agent-based Triton kernel generation across multiple hardware platforms.
Overview
Learn what KernelGenBench is, why it matters, and what it can do for you.
Features
Explore multi-source operators, multi-chip support, anti-hack validation, and evaluation metrics.
LLM Track
Evaluate LLMs on generating Triton kernels with Pass@K metric.
Agent Track
Evaluate coding agents that iteratively generate, verify, and optimize kernels.
Reference
Datasets, operators, hardware platforms, and technical specifications.
Development
Contributing guides, custom operators, and extending the framework.