iGEN
Visit IGEN World Explore IGEN Expo
EXPLORE UPGRADE PLANS
BREAKING
Moody's Assigns First-Time Baa2 Rating to RBL Bank, One Notch Above India's Sovereign Sebi Bars Zee's Subhash Chandra, Punit Goenka From Market for One Year Zepto Defers IPO by Two to Three Quarters After Tepid Investor Response Tim Cook: India Among Apple's Best Global Markets as June Quarter Records Revenue Domestic funds reach record 21% stake in Indian companies as FPI ownership drops to 17% Cybercriminals widen net as assessees rush to meet I-T return filing deadline Bloomberg Delays India's Sovereign Bond Index Inclusion as Market Reforms Need Further Testing Gold loans jump 93.8% y-o-y, fuel bank credit growth in Q1FY27 Snapchat joins YouTube, LinkedIn and Substack in fight against 'AI slop' Amazon speeds last-mile delivery, expands robotics fleet past 1 million Moody's Assigns First-Time Baa2 Rating to RBL Bank, One Notch Above India's Sovereign Sebi Bars Zee's Subhash Chandra, Punit Goenka From Market for One Year Zepto Defers IPO by Two to Three Quarters After Tepid Investor Response Tim Cook: India Among Apple's Best Global Markets as June Quarter Records Revenue Domestic funds reach record 21% stake in Indian companies as FPI ownership drops to 17% Cybercriminals widen net as assessees rush to meet I-T return filing deadline Bloomberg Delays India's Sovereign Bond Index Inclusion as Market Reforms Need Further Testing Gold loans jump 93.8% y-o-y, fuel bank credit growth in Q1FY27 Snapchat joins YouTube, LinkedIn and Substack in fight against 'AI slop' Amazon speeds last-mile delivery, expands robotics fleet past 1 million
Home ›› Technology ›› Ai ›› Llms ›› daVinci-kernel: Reinforcement Learning Framework Automates GPU Kernel Optimization with Co-Evolving Skill Library
iGEN
Technology Artificial Intelligence

daVinci-kernel: Reinforcement Learning Framework Automates GPU Kernel Optimization with Co-Evolving Skill Library

A new reinforcement learning framework called daVinci-kernel automates GPU kernel optimization by co-evolving skill selection, summarization, and utilization. The framework, detailed in a preprint on arXiv, uses three agents sharing one LLM backbone and achieves 37.2%, 70.6%, and 32.2% on KernelBench Level 1, 2, and 3 respectively, outperforming prior RL-trained models.

iG
iGEN Editorial
June 16, 2026
daVinci-kernel: Reinforcement Learning Framework Automates GPU Kernel Optimization with Co-Evolving Skill Library

Enterprises deploying large-scale AI and high-performance computing (HPC) workloads face a critical bottleneck: optimizing GPU kernels for peak performance. Manual kernel tuning is labor-intensive and requires deep expertise. A new reinforcement learning framework, daVinci-kernel, aims to automate this process by dynamically building and exploiting a skill library. According to a preprint on arXiv, daVinci-kernel jointly trains three agents in a co-evolution loop: a Skill Selection Agent, a Policy Agent, and a Skill Summary Agent.

How daVinci-kernel Works

DaVinci-kernel operates through a cooperative multi-agent system sharing a single LLM backbone. The Skill Selection Agent retrieves relevant techniques using BM25 and LLM reranking. The Policy Agent generates multi-turn CUDA or Triton kernels conditioned on the selected skills. The Skill Summary Agent distills successful rollouts into reusable skills. Crucially, candidate skills are only added after execution-based verification confirms reproducible speedups. The framework is initialized via a structured supervised fine-tuning (SFT) cold start on diversity-filtered data, then jointly optimized end-to-end with multi-turn REINFORCE and per-agent advantage estimation.

Performance on KernelBench

The researchers evaluated daVinci-kernel-14B on the KernelBench benchmark. Under the Fast$_1$ threshold, the model achieved:

Benchmark Level Success Rate
Level 1 37.2%
Level 2 70.6%
Level 3 32.2%

According to the preprint, daVinci-kernel-14B outperformed the strongest prior RL-trained model, this http URL-14B, across all three levels.

Implications for Enterprise AI Infrastructure

While currently focused on GPU kernel optimization, the daVinci-kernel framework demonstrates a generalizable approach to automated code optimization. For enterprise technology teams, this could reduce the time and expertise required to optimize CUDA and Triton kernels for specific hardware configurations. The use of a dynamically evolving skill library that is verified through execution ensures that only proven optimizations are retained. This approach could be extended to other domains where functional correctness is assumed but execution efficiency is the objective, such as database query optimization or network packet processing.

The preprint, authored by Fu, Dayuan, Jiang, Mohan, Wang, Tongyu, Yang, Dian, Hu, Jiarui, Liu, Liming, Hou, Jinlong, and Pengfei, is available on arXiv under the title "daVinci-kernel: Co-Evolving Skill Selection, Summarization, and Utilization via RL for GPU Kernel Optimization."

Sources:

Keep Reading

Recommended Stories

StarOR: New AI Framework Combines Tree Search and Reinforcement Learning for Optimization Modeling Technology

StarOR: New AI Framework Combines Tree Search and Reinforcement Learning for Optimization Modeling

A new AI framework called StarOR combines Monte Carlo Tree Search with test-time reinforcement learning to solve hierarchical optimization modeling problems. It decomposes modeling into four stages, uses a LoRA adapter updated via GRPO, and achieves state-of-the-art results on five benchmarks with a 4B parameter backbone, outperforming existing methods and frontier LLMs.

June 16, 2026
Reinforcement Learning Foundation Models: Synthetic MDPs Could Bridge the Gap Technology

Reinforcement Learning Foundation Models: Synthetic MDPs Could Bridge the Gap

The paper by Zighem, Abdelrahman, and Vie argues that reinforcement learning (RL) lacks a foundation model equivalent to those for language and vision. They propose using synthetic Markov Decision Processes (MDPs), which are as feasible to generate as synthetic tabular data, and demonstrate with a Graph Attention Network trained entirely on synthetic MDPs that achieves competitive results without task-specific tuning.

July 8, 2026
MENTOR: Reinforcement Learning via Flexible Teacher-Optimized Rewards for Tool-Use Distillation Technology

MENTOR: Reinforcement Learning via Flexible Teacher-Optimized Rewards for Tool-Use Distillation

A new research paper introduces MENTOR, a reinforcement learning framework that uses flexible teacher-optimized rewards to distill tool-use capabilities from large language models into small models. The approach improves out-of-domain generalization compared to supervised fine-tuning and strict reinforcement learning baselines.

July 8, 2026
Residual-Space Evolutionary Optimization via Flow-based Generative Models Technology

Residual-Space Evolutionary Optimization via Flow-based Generative Models

A new framework called residual-space evolutionary optimization addresses the challenge of data editing with non-differentiable objectives in flow-based generative models. By operating in residual space, it separates local exploitation (self-pollination) from broader exploration (cross-pollination). The method was validated on the MorphoMNIST benchmark and crystal data, showing balanced target alignment, instance preservation, and diversity.

June 22, 2026