MaxKernel: Agentic Kernel Generation for TPUs

类型：arxiv
标识：2609.04523
链接：https://arxiv.org/abs/2609.04523
主分类：agent
形态：method
TLDR：Designing and authoring high-performance custom kernels for accelerators is a complex task that requires deep hardware-level expertise. Large Language Models (LLM) can be leveraged together with real-time compiler feedback to build agentic systems for kernel generation. In this work, we present MaxKernel, a multi-agent system that implements three distinct paradigms for TPU kernel development: (1) a Human-in-the-Loop (HITL) agent for collaborative, step-by-step design; (2) an Autonomous (Auto) agent that executes a fully automated, metric/trace-driven optimization loop; and (3) a Graph-Based A
待LLM分类：否
来源文件：
/inbox/tom/_candidates/2026-09-07-agent-rag-longcontext-candidates.json

2026-08-11

笔记 Cool Papers · cs.CL (papers.cool) · RSS 摘要

2026-07-01

文档 Stephen 自测 · R57 · 2026-08-27

2026-08-27

笔记次精读：Argus — Uncertainty Quantification for Computer-Use Agents（短评）

2026-08-31

笔记 Tom 文献雷达 · AI Agent / RAG / 长上下文 · 2026-07-30（午后版）

2026-07-30

笔记 2026-07-09 研究笔记：RAG 优化与推理引擎实战 + Substack AI Agents Stack（**重写版 v2 · 2026-07-09 21:10 CST**）

2026-07-09

arXiv

2609.04523

TLDR

Designing and authoring high-performance custom kernels for accelerators is a complex task that requires deep hardware-level expertise. Large Language Models (LLM) can be leveraged together with real-time compiler feedback to build agentic systems for kernel generation. In this work, we present MaxKernel, a multi-agent system that implements three distinct paradigms for TPU kernel development: (1) a Human-in-the-Loop (HITL) agent for collaborative, step-by-step design; (2) an Autonomous (Auto) agent that executes a fully automated, metric/trace-driven optimization loop; and (3) a Graph-Based A

arXiv ↗ 中文全文

MaxKernel: Agentic Kernel Generation for TPUs

相关 · 按语义匹配