AgentKernelArena: Generalization-Aware Benchmarking of GPU Kernel Optimization Agents

Hugging Face Daily Papers Papers
gpu-kernel-optimization benchmark ai-coding-agents generalization open-source evaluation agent-workflows

Summary

AgentKernelArena is an open-source benchmark for evaluating AI coding agents on GPU kernel optimization, assessing full agent workflows and generalization to unseen configurations across 196 tasks.

GPU kernel optimization is increasingly critical for efficient deep learning systems, but writing high-performance kernels still requires substantial low-level expertise. Recent AI coding agents can iteratively read code, invoke compilers and profilers, and refine implementations, yet existing kernel benchmarks evaluate single LLM calls rather than full agent workflows, and none include both kernel-to-kernel optimization and unseen-configuration generalization testing. We present AgentKernelArena, an open-source benchmark for measuring AI coding agents on GPU kernel optimization. The benchmark contains 196 tasks spanning HIP-to-HIP optimization, Triton-to-Triton optimization, and PyTorch-to-HIP translation, and evaluates complete agent workflows in isolated workspaces using gated compilation, correctness, and performance checks, centralized scoring and an unseen-configuration generalization protocol that tests whether optimizations transfer to input configurations the agent never observed. Across production agents including Cursor Agent, Claude Code, and Codex Agent, we find near-perfect compilation and high correctness rates on most task categories, with the strongest configurations achieving mean speedups of up to 6.89x on PyTorch-to-HIP, 6.69x on HIP-to-HIP, and 2.13x on Triton-to-Triton tasks. Our unseen-configuration evaluation shows that HIP-to-HIP and Triton-to-Triton optimizations largely transfer to unseen input shapes, while PyTorch-to-HIP exhibits substantial correctness drops, indicating that agents generating kernels from scratch frequently hardcode shape-specific assumptions. AgentKernelArena is designed as a modular, extensible framework for rigorous evaluation of agentic GPU kernel optimization across agents, tasks, and hardware targets.
Original Article
View Cached Full Text

Cached at: 05/19/26, 06:30 AM

Paper page - AgentKernelArena: Generalization-Aware Benchmarking of GPU Kernel Optimization Agents

Source: https://huggingface.co/papers/2605.16819 Authors:

,

,

,

,

,

,

,

,

,

,

,

,

Abstract

AgentKernelArena is introduced as an open-source benchmark for evaluating AI coding agents on GPU kernel optimization, assessing full agent workflows and unseen-configuration generalization across multiple optimization tasks.

GPU kernel optimizationis increasingly critical for efficient deep learning systems, but writing high-performance kernels still requires substantial low-level expertise. RecentAI coding agentscan iteratively read code, invoke compilers and profilers, and refine implementations, yet existing kernel benchmarks evaluate single LLM calls rather than fullagent workflows, and none include both kernel-to-kernel optimization andunseen-configuration generalizationtesting. We present AgentKernelArena, an open-source benchmark for measuringAI coding agentsonGPU kernel optimization. The benchmark contains 196 tasks spanningHIP-to-HIP optimization,Triton-to-Triton optimization, andPyTorch-to-HIP translation, and evaluates completeagent workflowsin isolated workspaces using gatedcompilation,correctness, andperformance checks, centralized scoring and anunseen-configuration generalizationprotocol that tests whether optimizations transfer to input configurations the agent never observed. Across production agents including Cursor Agent, Claude Code, and Codex Agent, we find near-perfectcompilationand highcorrectnessrates on most task categories, with the strongest configurations achieving mean speedups of up to 6.89x on PyTorch-to-HIP, 6.69x on HIP-to-HIP, and 2.13x on Triton-to-Triton tasks. Our unseen-configuration evaluation shows that HIP-to-HIP andTriton-to-Triton optimizations largely transfer to unseen input shapes, while PyTorch-to-HIP exhibits substantialcorrectnessdrops, indicating that agents generating kernels from scratch frequently hardcode shape-specific assumptions. AgentKernelArena is designed as a modular, extensible framework for rigorous evaluation of agenticGPU kernel optimizationacross agents, tasks, and hardware targets.

View arXiv pageView PDFGitHub16Add to collection

Get this paper in your agent:

hf papers read 2605\.16819

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2605.16819 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.16819 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.16819 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles