Pith. sign in

REVIEW 16 cited by

Kevin: Multi-Turn RL for Generating CUDA Kernels

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2507.11948 v1 pith:SLACG3NT submitted 2025-07-16 cs.LG cs.AIcs.PFcs.SE

Kevin: Multi-Turn RL for Generating CUDA Kernels

classification cs.LG cs.AIcs.PFcs.SE
keywords kevincudakernelsmulti-turnacrosscorrectnessiterativelearning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

Writing GPU kernels is a challenging task and critical for AI systems' efficiency. It is also highly iterative: domain experts write code and improve performance through execution feedback. Moreover, it presents verifiable rewards like correctness and speedup, making it a natural environment to apply Reinforcement Learning (RL). To explicitly incorporate the iterative nature of this process into training, we develop a flexible multi-turn RL recipe that addresses unique challenges encountered in real-world settings, such as learning from long trajectories and effective reward attribution across turns. We present Kevin - K(ernel D)evin, the first model trained with multi-turn RL for CUDA kernel generation and optimization. In our evaluation setup, Kevin shows significant gains over its base model (QwQ-32B), improving correctness of generated kernels (in pure CUDA) from 56% to 82% and mean speedup from 0.53x to 1.10x of baseline (PyTorch Eager), and surpassing frontier models like o4-mini (0.78x). Finally, we study its behavior across test-time scaling axes: we found scaling serial refinement more beneficial than parallel sampling. In particular, when given more refinement turns, Kevin shows a higher rate of improvement.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 16 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. FastKernels: Benchmarking GPU Kernel Generation in Production

    cs.LG 2026-05 conditional novelty 8.0

    FastKernels is a production-aligned benchmark covering 96.2% of HuggingFace Transformers that reveals state-of-the-art kernel agents deliver at most 0.94x aggregate speedup.

  2. PassNet: Scaling Large Language Models for Graph Compiler Pass Generation

    cs.AI 2026-05 unverdicted novelty 7.0

    PassNet provides a dataset of 18K graphs and PassBench for LLM-generated compiler passes, with fine-tuned models achieving 2.67x gains on long-tail tasks where TorchInductor underperforms.

  3. Step-TP: A Grounded, Step-Level Dataset with Chain-of-Thought Reasoning for LLM-Guided Tensor Program Optimization

    cs.LG 2026-05 unverdicted novelty 7.0

    Step-TP is a dataset providing grounded, atomic step-level IR transitions and CoT supervision to enable reliable multi-step LLM-guided tensor program optimization instead of end-to-end imitation.

  4. CUDAHercules: Benchmarking Hardware-Aware Expert-level CUDA Optimization for LLMs

    cs.LG 2026-05 unverdicted novelty 7.0

    CUDAHercules benchmark demonstrates that leading LLMs generate functional CUDA code but fail to recover expert-level optimization strategies needed for peak performance on Ampere, Hopper, and Blackwell GPUs.

  5. CUDABeaver: Benchmarking LLM-Based Automated CUDA Debugging

    cs.LG 2026-05 unverdicted novelty 7.0

    CUDABeaver shows LLM CUDA debuggers often degenerate code for test-passing at the cost of speed, with protocol-aware metrics shifting success rates by up to 40 percentage points.

  6. Hawk: Harnessing Hardware-Aware Knowledge for High-Performance NPU Kernel Generation

    cs.AI 2026-07 conditional novelty 6.0

    Hawk raises NPU kernel generation accuracy from 49.4% to 80% and yields up to 2.2× speedups by retrieving and distilling structured hardware-aware knowledge without any model training.

  7. Hawk: Harnessing Hardware-Aware Knowledge for High-Performance NPU Kernel Generation

    cs.AI 2026-07 unverdicted novelty 6.0

    Hawk is a training-free framework that boosts NPU kernel generation accuracy to 80% and achieves up to 2.2x speedup via hardware-aware knowledge synthesis, 2D retrieval, and effect-driven distillation.

  8. SpecGen: Accelerating Agentic Kernel Optimization with Speculative Generation

    cs.DC 2026-06 unverdicted novelty 6.0

    SpecGen introduces speculative generation to fork non-reasoning kernel candidates during LLM reasoning traces, enabling early termination and parallel profiling to reduce end-to-end optimization time on H200 GPUs.

  9. Learning When to Optimize: Verified Optimization Skills from Expert GPU-Kernel Lineages

    cs.AI 2026-05 unverdicted novelty 6.0

    KLineage derives verified optimization skills from backward lineages of expert GPU kernels to guide LLM agents toward higher-quality and more efficient kernels than memory-based baselines.

  10. CUDABeaver: Benchmarking LLM-Based Automated CUDA Debugging

    cs.LG 2026-05 unverdicted novelty 6.0

    CUDABEAVER benchmark and pass@k(M,C,A) metric show LLM CUDA debugging success drops by up to 40 percentage points under strict performance requirements.

  11. Kernel Foundry: A Diagnosis-driven Evolutionary Kernel Optimizer with Multi-Experts

    cs.NE 2026-05 unverdicted novelty 6.0

    Kernel Foundry improves GPU kernel correctness and performance on KernelBench via expert-guided evolutionary search with diagnostic feedback and an experience library, reaching 100% correctness on Level 2.

  12. AdaExplore: Failure-Driven Adaptation and Diversity-Preserving Search for Efficient Kernel Generation

    cs.CL 2026-04 unverdicted novelty 6.0

    AdaExplore improves correctness and speed of Triton kernel generation by converting recurring failures into a memory of rules and organizing search as a tree that mixes local refinements with larger regenerations, yie...

  13. InCoder-32B-Thinking: Industrial Code World Model for Thinking

    cs.AR 2026-04 unverdicted novelty 6.0

    InCoder-32B-Thinking uses error-feedback synthesized thinking traces and a code world model to reach top open-source scores on general and industrial code benchmarks including 81.3% on LiveCodeBench and 84.0% on CAD-Coder.

  14. Kernel-Smith: A Unified Recipe for Evolutionary Kernel Optimization

    cs.CL 2026-03 unverdicted novelty 6.0

    Kernel-Smith combines evolutionary search with RL post-training to generate optimized GPU kernels, achieving SOTA speedups on KernelBench that beat Gemini-3.0-pro and Claude-4.6-opus on NVIDIA Triton and generalize to...

  15. The FIL Hypothesis: Inductive Biases Help with Kernel Engineering

    cs.AI 2026-06 unverdicted novelty 5.0

    The FIL Hypothesis claims that inductive biases outperform purely data-driven methods on GPU programming tasks with non-trivial feedback loops.

  16. AscendOptimizer: Episodic Agent for Ascend NPU Operator Optimization

    cs.LG 2026-03 unverdicted novelty 5.0

    AscendOptimizer combines kernel rewinding for reusable experience with evolutionary search on hardware feedback to optimize Ascend NPU operators, delivering 1.21x geometric-mean speedup and faster performance on 53.47...