Last updated: July 12, 2025
-
Go to QuickPod Templates
-
Find
Jupyter Lab CUDA 12.9, click Clone, and name it (e.g.,Gensyn 12.9)
-
In the Docker settings, add:
-p 3000:3000 --shm-size=16gThen click Save.
Set Disk Space to at least
50 -
Choose a machine:
- CPU: 16GB+ RAM
- GPU: 16GB+ RAM (VRAM can be less or more than 24GB)
Select the template you just created.
-
Click the orange Connect to JupyterLab button.
Inside JupyterLab, click the Terminal icon on the left.
Warning
Skip this if you don’t already have a swarm.pem file. If you do, drag and drop it into the file panel.
Paste and run the following.
git clone https://github.com/gensyn-ai/rl-swarm && apt update && apt install -y sudoIf you are prompted to overwrite the sudoers file, keep the current one by simply pressing ENTER.

Then, run these commands to update npm and node to the latest version.
sudo apt-get install -y curl
curl -fsSL https://deb.nodesource.com/setup_24.x | sudo -E bash -
sudo apt-get install -y nsolid
nsolid -v
npm install --force -g node@latestIn the terminal, run:
python3 -m venv .venv
source .venv/bin/activate
cd /workspace/rl-swarm && ./run_rl_swarm.shIf it gets stuck before the login page prompt loads, press CTRL+C, then run ./run_rl_swarm.sh again.
Click your port 3000 link. It will open a new browser tab.
Note
It may take 1–3 minutes. Just wait until you see this.
-
Enter the token sent to your inbox


If you are not receiving OTP email:
- OTP emails may be mistakenly filtered as spam. Look in your spam or junk folder for the OTP message
- Sometimes, there may be a delay in OTP delivery due to network or server issues. Wait a few minutes before requesting another OTP
- Double-check that the email address entered is correct and free of typos. An incorrect email will prevent delivery of the OTP
-
You should see Login Success
-
Back in the terminal, it will ask if you would like to upload your results to Huggingface. If you would, type "y", hit enter and then paste your Access Token and enter.

-
It will also ask if you would like to participate in the AI Prediction Market (Judge) functionality. If so, type "y" and press enter.

Warning
Note you may need to do step #5 multiple times until you don't get error. of Daemon failed to start in 15.0 seconds , Gensyn is quite unstable right now. I usually need to try 5 to 10 times on a bad day
If running is succesful you will see these, locate your ORG ID and Peer ID and remember it.
Go to: https://quickpod.github.io/gensyn
Note
Gensyn now uses a new swarm system called reasoning-gym instead of math problems. Old swarms are deprecated. New swarms will show 0 score at start.
Warning
Save swarm.pem locally to reconnect later with the same identity.
- Right-click
swarm.pem - Click Download
Note
Your node will stay running even if you close the browser. To check on it later, just reconnect to JupyterLab.
















