用可微分模型同时优化AI加速器硬件与算法映射，效率提升超12倍
DOSA: Differentiable Model-Based One-Loop Search for DNN Accelerators
- 构建可微分性能模型，用梯度下降联合搜索硬件参数与映射方案
- 在相同采样量下，比随机搜索快2.80倍，比贝叶斯优化快12.59倍
- 支持模块化扩展，可适配真实加速器，显著降低能时延积
在硬件设计空间探索中，同时优化硬件参数与算法到硬件的映射至关重要。以往方法将这两个高度非凸且庞大的空间分开探索，导致组合爆炸，优化困难。本文提出DOSA，结合可微分性能模型与基于梯度下降的优化技术，同步探索硬件设计空间与映射空间，快速定位高性能设计点。实验表明，在相似采样数量下，DOSA在改善DNN模型能时延积方面，分别优于随机搜索和贝叶斯优化2.80倍和12.59倍。我们还通过引入学习模型增强分析模型，实现了对真实DNN加速器的缓冲区大小与映射优化，能时延积降低1.82倍。
原文摘要 · Abstract (English)
In the hardware design space exploration process, it is critical to optimize both hardware parameters and algorithm-to-hardware mappings. Previous work has largely approached this simultaneous optimization problem by separately exploring the hardware design space and the mapspace - both individually large and highly nonconvex spaces - independently. The resulting combinatorial explosion has created significant difficulties for optimizers. In this paper, we introduce DOSA, which consists of differentiable performance models and a gradient descent-based optimization technique to simultaneously explore both spaces and identify high-performing design points. Experimental results demonstrate that DOSA outperforms random search and Bayesian optimization by 2.80x and 12.59x, respectively, in improving DNN model energy-delay product, given a similar number of samples. We also demonstrate the modularity and flexibility of DOSA by augmenting our analytical model with a learned model, allowing us to optimize buffer sizes and mappings of a real DNN accelerator and attain a 1.82x improvement in energy-delay product.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。