arXiv cs.CL· Yupeng Chen, Kunxiang Zhao, Xingjian Wang, Yue Zhao, Yijing Wu, Ye Liu, Kai Li, Congliang Chen, Ruoyu Sun·· 13 小时前AI 评分35
TrainBridge：用 GPU 编译证据指导 NPU 训练优化的智能体方法
TrainBridge: Agentic NPU Training Optimization Guided by GPU Compilation Evidence
AI 导读
TrainBridge 提出一种智能体方法，将成熟 GPU 编译栈产出的优化执行作为参考，提取优化意图来指导 Ascend NPU 训练程序重写，并通过迭代 profiling 精化。在覆盖 LLM 预训练、图像分类和视觉语言训练的 4 个负载上平均加速 1.7x，在 27 个 Modded-NanoGPT speedrun 记录上平均加速近 2.0x。
正文 · 原文
Abstract:We study how to improve training efficiency on Ascend NPUs. In the NPU training programs we study, framework-level compilation optimizations such as operator fusion are only partially applied, leaving fragmented computation and avoidable execution overheads. Profiling-guided tuning can address these inefficiencies, but it often requires costly trial and error. To make training on NPUs more efficient, we propose TrainBridge, an agentic approach for transferring GPU optimization guidance to NPU training. TrainBridge uses the highly optimized execution produced by mature GPU compilation stacks as a task-specific reference for NPU tuning. It extracts optimization intents from the GPU training execution to guide NPU program rewrites and further refines the rewritten program through iterative profiling. Our GPU-to-NPU migration experiments show that TrainBridge achieves an average 1.7x speedup across four workloads covering LLM pretraining, image classification, and vision-language training, and an average speedup of nearly 2.0x across 27 Modded-NanoGPT speedrun records with highly optimized GPU implementations.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.29171 [cs.CL] |
| (or arXiv:2610.29171v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.29171 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yupeng Chen [view email]
[v1]
Wed, 21 Oct 2026 04:00:48 UTC (655 KB)
来源：arXiv cs.CL · arxiv.org