🚀 RL-Swarm Setup Guide

Welcome to the RL-Swarm setup guide! This walkthrough helps you install and run the project from scratch, restore your key, activate the virtual environment, and expose your local service using tunnels.

---

🧭 Prerequisites

• ✅ swarm.pem key (backed up)
• ✅ Git, Python(3.10 min), screen, and npm installed
• ✅ GPU (3090, 4090, or ≥24GB VRAM recommended)
• ✅ CPU (arm64 or x86 CPU with minimum 32gb ram (note that if you run other applications during training it might crash training)

---

⚙️ Setup Steps

📁 1. Backup Your Key

Secure your existing swarm.pem key:

cp $HOME/rl-swarm/swarm.pem $HOME/

---

🧹 2. Clean & Clone Fresh Repo

Remove old project and clone fresh:

cd $HOME && \
rm -rf rl-swarm && \
git clone https://github.com/xailong-6969/rl-swarm.git

---

🔐 3. Restore swarm.pem

Place the key back into the new repo:

cp $HOME/swarm.pem $HOME/rl-swarm/

💡 Tip: Use your file explorer or ls to verify its placement.

---

🖥️ 4. Start a Screen Session

Start a persistent session:

cd $HOME/rl-swarm
screen -S gensyn

---

⚡ 5. Enable High-VRAM Optimization

For systems with ≥24GB VRAM (e.g. 3090/4090/A100/H100) (Note: only for gpu setups)

sed -i \
-e 's/use_vllm: false/use_vllm: true/' \
-e 's/fp16: false/fp16: true/' \
-e 's/gradient_checkpointing: false/gradient_checkpointing: true/' \
-e 's/num_train_samples: 2/num_train_samples: 1/' \
./rgym_exp/config/rg-swarm.yaml

⚡ Models:

• Gensyn/Qwen2.5-0.5B-Instruct
• Qwen/Qwen3-0.6B
• nvidia/AceInstruct-1.5B
• dnotitia/Smoothie-Qwen3-1.7B
• Gensyn/Qwen2.5-1.5B-Instruct

---

🐍 6. Set Up Python Environment

Inside the screen session:

python3 -m venv .venv
source .venv/bin/activate
./run_rl_swarm.sh

Wait for the message:

Waiting for localhost:3000...

Now, detach the screen:

Ctrl + A, then D

---

🌐 Expose Localhost Port 3000

Choose one of the methods below to make your service accessible online:

---

🚪 Option A: LocalTunnel

npm install -g localtunnel
lt --port 3000

---

☁️ Option B: Cloudflare Tunnel

sudo apt install cloudflared           # Ubuntu/Debian
# or
brew install cloudflared               # macOS

cloudflared tunnel --url http://localhost:3000

Open the URL in your browser to access your service.

---

🔄 Reconnect to Screen

To resume your running session:

screen -r gensyn

---

✅ You're Done!

Your RL-Swarm instance is now running, and you can interact with it either locally or via the exposed tunnel.