AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Kevin: Multi-Turn RL for Generating CUDA Kernels

Carlo Baronio, Pietro Marsella, Ben Pan, Simon Guo, Silas Alberti

ICLR 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train with multi-turn RL instead of single-turn RL (ablation)QwQ-32B (Kevin) · KernelBench eval, 16 traj x 8 turnsbest@16 speedup: 0.85x -> 1.10xImproved
RL training (multi-turn) on top of base modelQwQ-32B · KernelBench evalcorrectness best@16: 56% -> 82%Improved
more serial refinement turns at inference (16x8 vs 128x1) under fixed 128-kernel budget (ablation)Multi-Turn RL model · KernelBench evalpass@128 speedup: 0.65x -> 1.10xImproved
single-turn model with 128 parallel single-turn trajectories vs multi-turn model (ablation)Single-Turn RL · KernelBench eval, 128 traj x 1 turnpass@128 speedup: 0.70x vs 0.65xMixed
discounted sum reward with gamma 0.4 vs max aggregation or gamma 0.8 (ablation)Multi-turn RL reward · KernelBench, 16 traj x 8 turnsscaling with refinement turnsImproved
add KL penalty (0.001 or 0.01) to GRPO loss (ablation)GRPO training · KernelBench training rewardtraining rewardHurt
add constant length normalization and gradient norm clipping of 0.05 (ablation)GRPO training · KernelBench trainingtraining stability: junk onset delayed until step 100Improved
add a length penalty on the response (ablation)RL training (R1-Distill-Qwen-14B) · KernelBench trainingtraining behaviorHurt
set correctness weight to 1.0 or 0 in score (ablation)single-turn GRPO · KernelBench trainingtraining rewardHurt
reward intermediate objectives (compile or execute) (ablation)RL training · KernelBench trainingkernel correctnessHurt
use 64 parallel trajectories instead of 16 during multi-turn training (ablation)Kevin · KernelBench evalbest@16 correctness / performanceNo clear effect
use a stronger base model (QwQ-32B instead of R1-Distill-Qwen-7B) (ablation)RL training · KernelBench trainingreward hacking and learningImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy