facebookresearch/metaseq

Remove apex dependency after PyTorch 2.0?

Open

#617 创建于 2023年1月24日

在 GitHub 查看
 (3 评论) (0 反应) (1 负责人)Python (6,195 star) (701 fork)batch import
better-enggood first issue

描述

This is to look into whether or not we still need to use apex for speedups if out-of-the-box PyTorch 2.0 may "just work". Will require benchmarking at a few different scales to confirm.

贡献者指南

技术栈
pythonpytorch
领域
performance
议题类型
research
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。
3
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。
blocked
清晰度议题是否清楚说明期望改动、验收标准和下一步。
mostly clear
前置要求
PythonPyTorchapex librarybenchmarking
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。
15
研究方向
This issue investigates whether the apex library is still necessary with PyTorch 2.0. The investigation requires benchmarking training speed and memory usage at multiple scales (e.g., small, medium, large models). Relevant files include training scripts and configuration files in the repository. The assignee may have started work, so check recent branches or commits.