arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1806.01427cs.LGstat.ML
收藏

DAWNBench分析,一个时间准确的机器学习性能基准

Analysis of DAWNBench, a Time-to-Accuracy Machine Learning Performance Benchmark

Stanford DAWN(斯坦福大学DAWN实验室)

Cody Coleman, Daniel Kang, Deepak Narayanan, Luigi Nardi, Tian Zhao, Jian Zhang, Peter Bailis, Kunle Olukotun, Chris Re, Matei Zaharia

摘要:

研究人员提出了硬件、软件和算法优化来提高深度学习的计算性能。虽然这些优化中的一些更快地执行相同的操作(例如,增加GPU时钟速度),许多其他修改训练过程的语义(例如,精度降低),并可能影响最终模型对未知数据的准确性。由于缺乏考虑这些权衡的标准评估标准,很难直接比较这些优化。为了解决这个问题,我们最近推出了DAWNBench,这是一个专注于端到端训练时间的基准竞赛,以在一个看不见的数据集上实现接近最先进的准确性-一个称为时间准确性(TTA)的组合度量。在这项工作中,我们分析了来自DAWNBench的参赛作品,该作品收到了来自多个工业集团的优化提交,以调查TTA作为指标的行为以及表现最佳的参赛作品的趋势。我们表明,TTA具有较低的变异系数,并且为TTA优化的模型几乎与使用标准方法训练的模型一样通用。此外,尽管DAWNBench条目能够在3分钟内训练ImageNet模型,但我们发现它们仍然没有充分利用Tensor Cores等硬件功能。此外,我们发现,分布式条目可以花费超过一半的时间在通信上。我们在MLPERF v0.5基准测试的条目中显示了类似的结果。

英文摘要:

Researchers have proposed hardware, software, and algorithmic optimizations to improve the computational performance of deep learning. While some of these optimizations perform the same operations faster (e.g., increasing GPU clock speed), many others modify the semantics of the training procedure (e.g., reduced precision), and can impact the final model's accuracy on unseen data. Due to a lack of standard evaluation criteria that considers these trade-offs, it is difficult to directly compare these optimizations. To address this problem, we recently introduced DAWNBench, a benchmark competition focused on end-to-end training time to achieve near-state-of-the-art accuracy on an unseen dataset---a combined metric called time-to-accuracy (TTA). In this work, we analyze the entries from DAWNBench, which received optimized submissions from multiple industrial groups, to investigate the behavior of TTA as a metric as well as trends in the best-performing entries. We show that TTA has a low coefficient of variation and that models optimized for TTA generalize nearly as well as those trained using standard methods. Additionally, even though DAWNBench entries were able to train ImageNet models in under 3 minutes, we find they still underutilize hardware capabilities such as Tensor Cores. Furthermore, we find that distributed entries can spend more than half of their time on communication. We show similar findings with entries to the MLPERF v0.5 benchmark.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑