一句话：提出 Pruner 与 MoA-Pruner，以草拟-验证和在线适应机制加速张量程序调度搜索。
张量程序调优对深度神经网络的高效部署至关重要，但基于搜索的方法受复杂且学习缓慢的成本模型制约，常需数小时甚至数天才能找到高性能程序，且成本模型难以跨平台在线适应。本文提出 Pruner 与 MoA-Pruner。Pruner 采用“先草拟后验证”探索机制：先用基于符号的朴素分析器（草拟模型）筛选小规模潜在候选，再由学习成本模型识别最佳候选，避免对所有候选使用昂贵模型。MoA-Pruner 引入动量在线适应策略，缓解跨平台在线不适应问题。作者将 Pruner 集成到 TVM，在三个 GPU 平台上进行在线与离线调优实验，相比 Ansor、TenSet、TLP 和 MetaSchedule 取得最高约 4.82 倍的平均搜索加速。
Tensor program tuning is essential for the efficient deployment of deep neural networks. Search-based approaches have demonstrated scalability and effectiveness in automatically finding high-performance programs for specific hardware. However, the search process is often inefficient, taking hours or even days to discover optimal programs due to the exploration mechanisms guided by an accurate but slow-learned cost model. Meanwhile, the learned cost model trained on one platform cannot seamlessly adapt online to another, which we call cross-platform online unawareness. In this work, we propose Pruner and MoA-Pruner. Pruner is a ''Draft-then-Verify'' exploration mechanism that accelerates the schedule search process. Instead of applying the complex learned cost model to all explored candidates, Pruner drafts small-scale potential candidates by introducing a naive Symbol-based Analyzer (draft model), then identifies the best candidates by the learned cost model. MoA-Pruner introduces a Momentum online Adaptation strategy to address the cross-platform online unawareness. We incorporate Pruner into the TVM and conduct extensive experiments on three GPU-based platforms. Results show considerable speedup in schedule search time. In online tuning scenarios, Pruner and MoA-Pruner achieve an average speedup of 2.6 × and 4.82 × compared to Ansor. In offline tuning scenarios, Pruner achieves an average speedup of 4.75 × and 4.05× compared to TenSet and TLP, respectively. Furthermore, Pruner achieves an average speedup of 4.08 × compared to MetaSchedule on TensorCore.
@inproceedings{Qiao0HFZZZCATLY25,
title = {Pruner: A Draft-then-Verify Exploration Mechanism to Accelerate Tensor Program Tuning},
author = {Liang Qiao and Jun Shi and Xiaoyu Hao and Xi Fang and Sen Zhang and Minfan Zhao and Ziqi Zhu and Junshi Chen and Hong An and Xulong Tang and Bing Li and Honghui Yuan and Xinyang Wang},
booktitle = {ASPLOS},
year = {2025},
doi = {10.1145/3676641.3716269}
}