Towards Cold-Start Drafting and Continual Refining: A Value-Driven Memory Approach with Application to NPU Kernel Synthesis

• URL: http://arxiv.org/abs/2603.10846v1
• Date: Wed, 11 Mar 2026 14:57:06 GMT
• Title: Towards Cold-Start Drafting and Continual Refining: A Value-Driven Memory Approach with Application to NPU Kernel Synthesis
• Abstract summary: EvoKernel is a self-evolving agentic framework that automates the lifecycle of kernel synthesis.<n>It learns stage-specific Q-values that prioritize experiences based on their contribution to the current objective.<n>It improves models' correctness from 11.0% to 83.0% and a median speedup of 3.60x over initial drafts.
• Score: 68.7701048879757
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Deploying Large Language Models to data-scarce programming domains poses significant challenges, particularly for kernel synthesis on emerging Domain-Specific Architectures where a "Data Wall" limits available training data. While models excel on data-rich platforms like CUDA, they suffer catastrophic performance drops on data-scarce ecosystems such as NPU programming. To overcome this cold-start barrier without expensive fine-tuning, we introduce EvoKernel, a self-evolving agentic framework that automates the lifecycle of kernel synthesis from initial drafting to continual refining. EvoKernel addresses this by formulating the synthesis process as a memory-based reinforcement learning task. Through a novel value-driven retrieval mechanism, it learns stage-specific Q-values that prioritize experiences based on their contribution to the current objective, whether bootstrapping a feasible draft or iteratively refining latency. Furthermore, by enabling cross-task memory sharing, the agent generalizes insights from simple to complex operators. By building an NPU variant of KernelBench and evaluating on it, EvoKernel improves frontier models' correctness from 11.0% to 83.0% and achieves a median speedup of 3.60x over initial drafts through iterative refinement. This demonstrates that value-guided experience accumulation allows general-purpose models to master the kernel synthesis task on niche hardware ecosystems. Our official page is available at https://evokernel.zhuo.li.

Related papers

• Accelerating Unified Multimodal Models with Core-Expansion Routing and Unified Computation Scheduling [61.28599393707943]
  Unified multimodal models jointly support understanding and generation, but incur substantial redundant computation across tokens, layers, and generation timesteps.<n>We propose a task-shared core scorer and progress-conditioned generation expansions, optimized through generation decomposition and cross-task core alignment.<n> Experiments on two representative UMM architectures demonstrate consistent quality--efficiency improvements across both tasks, retaining 98.03% of dense understanding performance with a 1.93times$ end-to-end inference speedup.
  arXiv  Detail & Related papers  (2026-08-29T14:27:55Z)
• Hawk: Harnessing Hardware-Aware Knowledge for High-Performance NPU Kernel Generation [11.866479609153194]
  We introduce Hawk, a training-free framework that harnesses hardware-aware knowledge through three core modules.<n>Hawk elevates generation accuracy from 49.4% to 80.0%, while achieving up to a 2.2x execution speedup over state-of-the-art baselines.
  arXiv  Detail & Related papers  (2026-07-02T01:41:34Z)
• Nimbus: A Unified Embodied Synthetic Data Generation Framework [51.55989844555466]
  Scaling data volume and diversity is critical for generalizing embodied intelligence.<n>We present Nimbus, a unified synthetic data generation framework designed to integrate heterogeneous navigation and manipulation pipelines.<n>Our evaluation demonstrates that Nimbus achieves a 2-3X improvement in end-to-end throughput compared to unoptimized baselines.
  arXiv  Detail & Related papers  (2026-01-29T09:27:31Z)
• Towards Automated Kernel Generation in the Era of LLMs [17.69471168609145]
  Kernel engineering is a time-consuming and non-scalable process.<n>Recent advances in large language models (LLMs) and agentic systems have opened new possibilities for automating kernel generation and optimization.<n>The field remains fragmented, lacking a systematic perspective for LLM-driven kernel generation.
  arXiv  Detail & Related papers  (2026-01-22T07:53:52Z)
• AscendKernelGen: A Systematic Study of LLM-Based Kernel Generation for Neural Processing Units [39.846358001824996]
  We propose Ascend KernelGen, a generation-evaluation integrated framework for NPU kernel development.<n>We introduce Ascend-CoT, a high-quality dataset incorporating chain-of-thought reasoning derived from real-world kernel implementations.<n>We also design NPU KernelBench, a comprehensive benchmark for assessing compilation, correctness, and performance across varying complexity levels.
  arXiv  Detail & Related papers  (2026-01-12T03:12:58Z)
• From RISC-V Cores to Neuromorphic Arrays: A Tutorial on Building Scalable Digital Neuromorphic Processors [0.37863160479533026]
  Digital neuromorphic processors are emerging as a promising computing substrate for low-power, always-on EdgeAI applications.<n>This tutorial paper outlines the main architectural design principles behind fully digital neuromorphic processors.
  arXiv  Detail & Related papers  (2025-11-27T14:14:23Z)
• ConCuR: Conciseness Makes State-of-the-Art Kernel Generation [5.010229074860956]
  Key challenge for kernel generation is the scarcity of high-quality data.<n>We develop a pipeline that generates and curates high-quality kernels with reasoning traces.<n>We show that the average reasoning length can serve as a metric to assess the difficulty of kernel generation tasks.
  arXiv  Detail & Related papers  (2025-10-08T15:41:15Z)
• TD3: Tucker Decomposition Based Dataset Distillation Method for Sequential Recommendation [50.23504065567638]
  This paper introduces textbfTD3, a novel textbfDataset textbfDistillation method within a meta-learning framework.<n> TD3 distills a fully expressive emphsynthetic sequence summary from original data.<n>An augmentation technique allows the learner to closely fit the synthetic summary, ensuring an accurate update of it in the emphouter-loop.
  arXiv  Detail & Related papers  (2025-02-05T03:13:25Z)
• Boosting Low-Data Instance Segmentation by Unsupervised Pre-training with Saliency Prompt [103.58323875748427]
  This work offers a novel unsupervised pre-training solution for low-data regimes. Inspired by the recent success of the Prompting technique, we introduce a new pre-training method that boosts QEIS models. Experimental results show that our method significantly boosts several QEIS models on three datasets.
  arXiv  Detail & Related papers  (2023-02-02T15:49:03Z)
• Neural Attentive Circuits [93.95502541529115]
  We introduce a general purpose, yet modular neural architecture called Neural Attentive Circuits (NACs) NACs learn the parameterization and a sparse connectivity of neural modules without using domain knowledge. NACs achieve an 8x speedup at inference time while losing less than 3% performance.
  arXiv  Detail & Related papers  (2022-10-14T18:00:07Z)
• Kernel Identification Through Transformers [54.3795894579111]
  Kernel selection plays a central role in determining the performance of Gaussian Process (GP) models. This work addresses the challenge of constructing custom kernel functions for high-dimensional GP regression models. We introduce a novel approach named KITT: Kernel Identification Through Transformers.
  arXiv  Detail & Related papers  (2021-06-15T14:32:38Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.