genv 0.9.0

GPU Environment Management

Genv (GPU Environment Management)

Genv lets you easily control, configure and monitor the GPU resources that you are using.

It is intendend to ease up the process of GPU allocation for data scientists without code changes 💪🏻

This project was highly inspired by pyenv and other version, package and environment management software like Conda, nvm, rbenv.

🔥 Why Genv?

• Easily share GPUs with your teammates
• Find available GPUs for you to use - on-prem or on cloud via remote access
• Pool the GPUs from multiple machines, allocate the available machine without SSH-ing every one of them
• Switch between GPUs without code changes
• Reserve GPU resources for as long as you use them with no one else hijacking them
• Reproduce your experiment environment easily
• Save time while collaborating

Plus, it's 100% free and gets installed before you can say Jack Robinson.

Documentation

Check out the Genv documentation site.

🏃🏻 Join us in the AI Infrastructure Club!

Looking for a place to discuss best practices, discover new tools, and exchange ideas about how to make the most out of our GPUs without losing time? Join the Discord server with the creators of Genv and rntop - start building your models faster!

• Installation and setup support as well as best practice tips and tricks directly for your use-case
• Discuss possible features
• Monthly Beers with Engineers sessions with amazing guests
• Networking events
• and many more...

🎉 Simple Integration & Usage with your fav IDE

Integration with VSCode (Take me to the installation guide!)

Integration with JupyterLab (Take me to the installation guide!)

A PyCharm integration is also in our roadmap so stay tuned!

License

The Genv software is Copyright 2022 [Run.ai Labs, Ltd.]. The software is licensed by Run.ai under the AGPLv3 license. Please note that Run.ai’s intention in licensing the software are that the obligations of licensee pursuant to the AGPLv3 license should be interpreted broadly. For example, Run.ai’s intention is that the terms “work based on the Program” in Section 0 of the AGPLv3 license, and “Corresponding Source” in Section 1 of the AGPLv3 license, should be interpreted as broadly as possible to the extent permitted under applicable law.

Genv Remote vs. Run.ai

1. containers vs environments
2. SSH vs k8s
3. interception- vs monitoring-based enfircements
4. non-centralized (no load balancing)
5. no advanced user roles (everyone can list envs and devices)
6. shim-based visualization vs. interception-based virtualization
7. environment must fit in a node vs. multi-node workload
8. no compute mgmt

QUOTA:

system: [V] (nvidia-smi) non env processes [V] (nvidia-smi) processes from envs on non-attached devices [ ] (envs.json) timeout

resources: [V] (devices.json) num of attached devices

(later...)
[ ] (nvidia-smi)      num of processes (as an alternative to compute mgmt)
[ ] (envs.json)       num of active envs
[ ] (envs.json)       num of machines with active envs
[ ] (nvidia-smi)      amount of GPU memory (per device/total)

https://grafana.com/docs/grafana/latest/cli/#override-homepath-value