AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Dr. Kernel: Reinforcement Learning Done Right for Triton Kernel Generations

Wei Liu, Jiawei Xu, Yingru Li, Longtao Zheng, Tianjian Li, Qian Liu, Junxian He

ICML 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 0 citations (OpenAlex)

Read the paper

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
replace GRPO mean baseline with TRLOO leave-one-out baseline (ablation)multi-turn RL for Triton kernels · KernelBench Level 2, Qwen3-8B-BaseFast@1Improved
train multi-turn instead of single-turn (ablation)RL with TRLOO · KernelBench Level 2, Qwen3-8B-BaseFast@1Improved
use reward-to-go return (gamma=1) instead of gamma=0 (ablation)TRLOO multi-turn RL · KernelBench Level 2Fast@1 first turnImproved
add execution-based hacking check to the environment (ablation)KernelGYM RL training · KernelBench Level 2 validationFast@1 under hacking checkImproved
add mismatch rejection sampling (MRS) (ablation)TRLOO multi-turn RL · KernelBench Level 2training stability and Fast@1.2Mixed
add profiling-based rewards (PR) and profiling-based rejection sampling (PRS) on top of MRS (ablation)TRLOO multi-turn RL · KernelBench Level 2Fast@1.2Improved
multi-turn RL after cold-start SFTQwen3-8B-Base · KernelBench Level 2Fast@1.2: 5.6 -> 20.0Improved
add sequential test-time scaling with context managementDr. Kernel-14B · KernelBench Level 2Fast@1.2: 25.6 -> 31.6Improved
select best turn across history instead of last turnDr. Kernel-14B-STTS · KernelBench Level 2Fast@1.2: 31.6 -> 47.8Improved
context management (top-4 turns by reward) instead of vanilla extrapolation (ablation)test-time scaling · KernelBench Level 2Fast@1.2 over turnsMixed
evaluate under torch.compile baselineDr. Kernel-14B · KernelBench Level 2Fast@1.2: 25.6 -> 1.9Hurt
scale to Level 3 with the same trainingDr. Kernel-14B · KernelBench Level 3Fast@1.2: 1.2 vs GPT-5 12.0No clear effect

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