WangmerlynWangmerlynCommunityยท1 Agent Skills Included

KeepGPU

Keep shared GPUs alive with lightweight VRAM sessions

Keeps CUDA, ROCm, and Apple Silicon GPUs active on shared machines by holding a small VRAM signal. Prevents schedulers and idle timeouts from releasing your GPU between workloads without manual babysitting. Offers blocking and background service modes with start, status, and stop commands plus a live dashboard. Backs off automatically when GPU utilization is high so real workloads are never disturbed.
npx skills add Wangmerlyn/KeepGPU --all -g -y

All Skills in This Repository (1)

Pure Emerald Level Indicators

Frequently Asked Questions

FAQPage Schema
How to install KeepGPU?โ–ผ

Run `npx skills add Wangmerlyn/KeepGPU --all -g -y` in your terminal to install the skill globally. You also need PyTorch and the keep-gpu Python package installed for the underlying CLI to work.

How to stop a GPU from being released on shared servers?โ–ผ

Run `keep-gpu start --gpu-ids 0 --vram 1GiB --interval 60` to hold a lightweight VRAM signal that keeps the GPU active. Stop it anytime with `keep-gpu stop --all`.

Does KeepGPU work on Mac and AMD GPUs?โ–ผ

Yes. It supports NVIDIA CUDA, AMD ROCm/HIP, and Apple Silicon MPS, as long as PyTorch can see the device on your machine.

Will KeepGPU slow down other users on a shared machine?โ–ผ

No. It monitors GPU utilization and automatically backs off when the card is busy, only running keepalive cycles when the GPU is idle.

Can I monitor my KeepGPU sessions in a browser?โ–ผ

Yes. KeepGPU ships with a local web dashboard at http://127.0.0.1:8765/ where you can view active sessions and GPU status.

Related Repositories in Software Engineering

View All in Software Engineeringโ†’