AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

CUDA-L1: Improving CUDA Optimization via Contrastive Reinforcement Learning

Xiaoya Li, Xiaofei Sun, Albert Wang, Jiwei Li, Chris Shum

ICLR 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models · 58 citations (OpenAlex)

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
replace contrastive RL with vanilla GRPO after stage 1+2 (ablation)CUDA-L1 · KernelBench all levels (A100)mean speedup: 2.41 -> 3.12Improved
add contrastive RL stage after SFT and self-supervised stages (ablation)CUDA-L1 · KernelBench all levels (A100)mean speedup: 1.36 -> 3.12Improved
add self-supervised learning stage after SFT (ablation)CUDA-L1 · KernelBench all levels (A100)mean speedup: 1.14 -> 1.36Improved
use bucket or island exemplar sampling instead of random sampling (ablation)contrastive RL · KernelBench all levels (A100)mean speedup: 2.14 -> 3.12Improved
island sampling versus bucket sampling (ablation)contrastive RL · KernelBench all levels (A100)mean speedup: 3.21 vs 3.12No clear effect
update model parameters (RL) instead of fixed-parameter evolutionary prompting (ablation)DeepSeek-R1 / CUDA-L1 · KernelBench all levels (A100)mean speedup: 1.41 -> 3.12Improved
add evolutionary contrastive prompting to vanilla foundation models (ablation)DeepSeek-R1 · KernelBench all levels (A100)mean speedup: 0.88 -> 1.41Improved
apply A100-trained kernels to other GPUsCUDA-L1 · KernelBench all levelsmean speedup: 2.38x to 3.85x across H20, 3090, L40, H100Improved
compare against torch.compile baselines instead of default on Level 3CUDA-L1 · KernelBench Level 3 (A100)mean speedup: 2.96 -> 1.62Hurt
add CUDA Graphs to LSTM kernel optimization (ablation)CUDA-L1 LSTM kernel · KernelBench Level 3 Task 35speedup: 1.0x -> 2.77x-3.42xImproved
add mathematical short-circuit to Conv3d task (ablation)CUDA-L1 kernel · KernelBench Level 2 Task 38speedup: 1.0x -> 28.6x+Improved

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