pytorch/ignite

Tracking the metrics with performance gaps on XLA

Open

#2,109 创建于 2021年7月14日

在 GitHub 查看
 (2 评论) (0 反应) (0 负责人)Python (4,313 star) (602 fork)batch import
HacktoberfestPyDataGlobalenhancementhelp wantedmodule: metrics

描述

🚀 Feature

Since this commit, the metrics have been disabled with xla device due to a performance regression.

https://github.com/pytorch/ignite/blob/7bdf92322a15ba0bea71ad76f3f785b7ea5907c3/ignite/metrics/metric.py#L224-L225

To solve this, it would be interesting to track which metrics work fine and which have lack of performance. Thus, we could think about more specific enable / disable process.

贡献者指南

技术栈
python
领域
performancemachine learning
议题类型
feature
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
4
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
1-2 days
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
stale
清晰度议题是否清楚说明期望改动、验收标准和下一步。
needs investigation
前置要求
无
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
25
研究方向
Investigate the referenced commit 002b595daa8a8345286c5e096c33e278948686a7 to understand why metrics were disabled on XLA. Examine the code in ignite/metrics/metric.py lines 224-225 to identify the performance regression. Then, measure which metrics (e.g., Accuracy, Loss) perform well on XLA and which cause slowdowns. The goal is to propose a more granular enabling/disabling mechanism for metrics on XLA devices.