Awesome AI Kernel Engineering

Methods

Title	Venue	Date	Code	topic
KFORGE: PROGRAM SYNTHESIS FOR DIVERSE AI HARDWARE ACCELERATORS	arXiv	2025.11		CUDA
Towards Robust Agentic CUDA Kernel Benchmarking, Verification, and Optimization	arXiv	2025.09	Github	CUDA
Autonomous Code Evolution Meets NP-Completeness	arXiv	2025.09	-	Nvidia Official
Astra: A Multi-Agent System for GPU Kernel Performance Optimization	arXiv	2025.09	-	CUDA
Towards Agentic OS: An LLM Agent Framework for Linux Schedulers	arXiv	2025.09	-	General Kernel, OS Schedule
SwizzlePerf: Hardware-Aware LLMs for GPU Kernel Performance Optimization	arXiv	2025.08	-	AMD GPU
CudaLLM: Training Language Models to Generate High-Performance CUDA Kernels	HugginFace	2025.08	Github	CUDA, Finetuning
OS-R1: Agentic Operating System Kernel Tuning with Reinforcement Learning	arXiv	2025.08	-	General Kernel, OS Kernel
TimelyHLS: LLM-Based Timing-Aware and Architecture-Specific FPGA HLS Optimization	arXiv	2025.07	-	Hardware_as_Code, HLS
GEAK: Introducing Triton Kernel AI Agent & Evaluation Benchmarks	arXiv	2025.07	Github	Triton (AMD), Iterative Search
GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning	arXiv	2025.07	-	AMD NPU + CUDA, Prompt Engineering
CUDA-L1: Improving CUDA Optimization via Contrastive Reinforcement Learning	arXiv	2025.07	Github	CUDA, Finetuning
Omniwise: Predicting GPU Kernels Performance with LLMs	arXiv	2025.06	-	AMD NPU, Surrogate Modeling
Kevin: Multi-Turn RL for Generating CUDA Kernels	EXAIT Workshop @ ICML	2025.06	-	CUDA, Finetuning
GPU Kernel Scientist: An LLM-Driven Framework for Iterative Kernel Optimization	ES-FoMo Workshop @ ICML	2025.06	-	AMD, Iterative Search
AlphaEvolve: A coding agent for scientific and algorithmic discovery	arXiv	2025.06	Github1	TPU, Agentic System
CUDA-LLM: LLMs Can Write Efficient CUDA Kernels	arXiv	2025.06	-	CUDA, Prompt Engineering (?)
The AI CUDA Engineer: Agentic CUDA Kernel Discovery, Optimization and Composition	arXiv	2025.02	HuggingFace	CUDA, Agent + Iterative Search + RAG
ECCO: Can We Improve Model-Generated Code Efficiency Without Sacrificing Functional Correctness? (Waghjale et al., EMNLP 2024)	EMNLP 2024	2024.11	-	LLM4Code

Datasets and Benchmarks

Title	Venue	Date	Code	topic
MultiKernelBench: A Multi-Platform Benchmark for Kernel Generation	arXiv	2025.07	Github	CUDA + CANN + TPU
NPUEval: Optimizing NPU Kernels with LLMs and Open Source Compilers	arXiv	2025.07	-	NPU
TritonBench: Benchmarking Large Language Model Capabilities for Generating Triton Operators	ICML	2025.02	Github	Triton (CUDA)
KernelBench: Can LLMs Write Efficient GPU Kernels?	ICML	2025.02	Github	CUDA
Comparing Llama-2 and GPT-3 LLMs for HPC kernels generation	LCPC	2023.09	-

Others

Title	Date	topic
AMD Developer Challenge 2025	2025.04	AMD GPU
GPU Mode Learderboard	-	AMD + CUDA

Footnotes

1. A public implementation of AlphaEvolve. ↩