google-deepmind / google-deepmind/dm_control

dm-control mujoco tasks use A LOT of cpu's

未关闭
#425 4 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看
主要语言
Python
星标
4.7k
派生
764
PR 合并指标
30 天内没有已合并 PR

描述

[Potentially related to this post](https://github.com/google-deepmind/mujoco/issues/203). I was wondering why my code slows down when running a few seeds at a time. When I looked, the CPU's are being throttled by running multiple instances of the MuJoCo environments (specifically I was checking with walker-walk). Using just one environment was maxing out several cores. Is there a way to stop this? I compared with hopper/half-cheetah from the Gymnasium implementations and they use just a single CPU per environment. Do I need to change some setting in dm-control to fix this?

If any more information is needed (e.g. resource usage from the terminal), then please let me know what else I can add.

贡献指南

打开贡献指南

调研方向

首先,使用 dm-control 的 walker-walk 环境和多个 seed 重现已报告的 CPU 使用情况,然后将其与 Gymnasium 的 hopper 和 half-cheetah 实现进行比较。调查 dm-control 是否提供了用于控制每个环境 CPU 使用量的设置。完成的标准是确定原因,并记录或解决防止意外使用多核 CPU 的方法。

由索引模型根据 Issue 内容生成。

评估

技术栈
python
领域
machine-learning, performance
Issue 类型
缺陷
难度
4/5
预计耗时
3-5 天
活跃度
停滞
描述清晰度
需要澄清
新手友好度
25/100

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。