contribute
Train a model with us
Agora runs are open to anyone with a suitable GPU. Your machine hosts one stage of the model and trains alongside other contributors over the internet. It can join late and leave at any time; the run carries on.
What you needFull requirements
- GPU24 GB of VRAM or more: RTX 4090, RTX 5090, RTX 6000 Ada, L40S
- Memory70 GB of RAM per GPU and 80 GB of free disk
- Network200 Mbps down, 100 Mbps up, and one open TCP port
- LocationNorth America for now, within 80 ms of the other peers
- SystemLinux, or Windows with WSL2
- IdentityA Hugging Face account. There is no other sign-up.
No machine that fits? Spot and pre-emptible instances work, and the cloud guide covers the cheapest options on each provider.
How it works
Check that a run is open
Runs are announced on the dashboard, which also shows the current wait to join.
Start a node
Clone the client and run it. The CLI walks you through the rest.
git clone https://github.com/PluralisResearch/agora-test cd agora-test python3 agora_cli.py
Watch it train
Your node appears on the public leaderboard within minutes. Points track the compute you contribute and the hours you stay online. They carry no financial reward at this stage.
Go further
- Quick start guide
- Renting a GPU: cloud and marketplace options
- How Agora works
- Ask for help on Zulip
- Join the community
Why any of this matters is on the mission page.