mlsanityData to 7 Oct 2026

Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation

arXiv 2605.26720 ↗First posted 26 May 2026cs.AIcited by 0

On the map: LLM agents and tool use → General LLM agent frameworks and benchmarks

also in 1 more
Abstract (arXiv)

Large language models (LLMs) have shown strong empirical gains as self-evolving agents for CUDA kernel generation, driven by feedback-conditioned planning across generations. However, how planning decisions attribute and combine heterogeneous feedback signals remains opaque. Standard end-to-end ablations fail to resolve this question, as iterative planning amplifies early perturbations and conflates feedback effects with trajectory-dependent drift. We introduce textttCUDAnalyst, a unified analysis layer for controlled, generation-level attribution of planning decisions to feedback components via trajectory freezing and selective feedback injection. textttCUDAnalyst enables stable generation-level evaluation and principled coalitional-style attribution of feedback effects and interactions. Our results show that explicit planning is beneficial only when feedback is aligned, that effective planning emerges from structured multi-feedback interactions, and that high-level plans from stronger reasoning models can partially transfer to weaker ones. These trends hold across reference backbones, representative workloads, and reference induction regimes, indicating that the identified feedback-to-plan structure is robust within the controlled axes studied.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use. Builds on 28 of its 57 references that are in the map. First posted on arXiv 26 May 2026.

Builds on

28 of 57 references.

Closest ancestors · 1

Same direction, newest first

  1. 2025Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents29 May

Other references · 27

  1. 2021Evaluating Large Language Models Trained on Codecited by 8,053
  2. 2023Towards Best Practices of Activation Patching in Language Models: Metrics and Methodscited by 328
  3. 2024Evolution of Heuristics: Towards Efficient Automatic Algorithm Design Using Large Language Modelcited by 258
  4. 2025KernelBench: Can LLMs Write Efficient GPU Kernels?cited by 206
  5. 2025ShinkaEvolve: Towards Open-Ended And Sample-Efficient Program Evolutioncited by 146
  6. 2025Monte Carlo Tree Search for Comprehensive Exploration in LLM-Based Automatic Heuristic Designcited by 100
  7. 2014Propagation Kernelscited by 88
  8. 2025CUDA-L1: Improving CUDA Optimization via Contrastive Reinforcement Learningcited by 59
  9. 2018GraKeL: A Graph Kernel Library in Pythoncited by 54
  10. 2018Tiramisu: A Polyhedral Compiler for Expressing Fast and Portable Codecited by 53
  11. 2025Towards Robust Agentic CUDA Kernel Benchmarking, Verification, and Optimizationcited by 52
  12. 2025Astra: A Multi-Agent System for GPU Kernel Performance Optimizationcited by 46
15 more
  1. 2025Reducing Cost of LLM Agents with Trajectory Reductioncited by 44
  2. 2025CudaForge: An Agent Framework with Hardware Feedback for CUDA Kernel Optimizationcited by 42
  3. 2024LLM4AD: A Platform for Algorithm Design with Large Language Modelcited by 33
  4. 2025STARK: Strategic Team of Agents for Refining Kernelscited by 23
  5. 2026FlashInfer-Bench: Building the Virtuous Cycle for AI-driven LLM Systemscited by 20
  6. 2025Interpreting Emergent Planning in Model-Free Reinforcement Learningcited by 15
  7. 2025ConCuR: Conciseness Makes State-of-the-Art Kernel Generationcited by 13
  8. 2025SwizzlePerf: Hardware-Aware LLMs for GPU Kernel Performance Optimizationcited by 11
  9. 2025PRAGMA: A Profiling-Reasoned Multi-Agent Framework for Automatic Kernel Optimizationcited by 11
  10. 2023Prompt Valuation Based on Shapley Valuescited by 10
  11. 2024AgentMonitor: A Plug-and-Play Framework for Predictive and Secure Multi-Agent Systemscited by 9
  12. 2025Understanding and Optimizing Agentic Workflows via Shapley valuecited by 9
  13. 2025cuPilot: A Strategy-Coordinated Multi-agent Framework for CUDA Kernel Evolutioncited by 7
  14. 2025Agentic Auto-Scheduling: An Experimental Study of LLM-Guided Loop Optimizationcited by 4
  15. 2025Optimizing PyTorch Inference with LLM-Based Multi-Agent Systemscited by 4

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2023AdaPlanner: Adaptive Planning from Feedback with Language Models
  2. 2025Deliberate Planning in Language Models with Symbolic Representation
  3. 2026GRASP: Generating, Revising, and Assessing for Strategic Planning with Agentic AI
  4. 2025SPIRAL: Symbolic LLM Planning via Grounded and Reflective Search
  5. 2026Model Space Reasoning as Search in Feedback Space for Planning Domain Generation
  6. 2026From Coarse to Fine: Self-Adaptive Hierarchical Planning for LLM Agents
  7. 2025Learning When to Plan: Efficiently Allocating Test-Time Compute for LLM Agents
  8. 2023Improving Planning with Large Language Models: A Modular Agentic Architecture