A Comprehensive Guide to Running a Gensyn RL-Swarm Training
Continuous Training, an Experimental (advanced) mode, GGUF LLM Compatibility on Huggingface

Our pick an experimental (advanced) mode at this model a continuously trained Qwen2.5-Coder-0.5B-Instruct fine-tuned using Gensyn RL-Swarm framework with GRPO (Group Relative Policy Optimization) and supported format GGUF (llama.cpp) for enhanced code generation capabilities. Note: Current training focuses on programming challenges with adaptive weighted sampling.

• Agent ID: Huggingface LLModels
• Training Status: 🟢 LIVE - Model updates automatically every 5-10 minutes
• Auto-Sync GGUF Pipeline Status: 🟢 LIVE - Commits update automatically every hourly
• Current Progress: Round 14,854+ / 100,000 (14.85%)
• Framework Version: Gensyn RL-Swarm (CodeZero) v0.7.0
• Contract Judge: SwarmCoordinator v0.4.2
• GGUF Quantization: Multiple quantized available (F16, Q3_K_M, Q4_K_M, Q5_K_M, Q6_K)

• Proposers: Generate coding problems and unit tests, adjusting difficulty dynamically based on solver performance. Proposers create challenges that adapt to the swarm's current capabilities, ensuring continuous learning opportunities.
• Solvers: Attempt coding challenges, learn locally through RL, and share rollouts with peers. Solvers exchange solutions to promote diversity and accelerate collective learning across the network.
• Evaluators: Frozen models that assess correctness and assign rewards. Evaluators use rule-based assessment to score submissions without executing code, ensuring safety and scalability.

Note: Its just a imagine, you can choose anything your take. But i can sharing any tips if your rent cloud GPU to https://octa.space I'm not promoter choose this my principals rent any low-cost, its very cheaper than rental GPU competitors. For tips & trick read... https://github.com/arcxteam/octa-rental-gpu

Basically, if you rent a GPU with a minimum of 6GB or even 8-12-16-24GB of VRAM, you can run other nodes because only 4-5GB of VRAM will be used for this GENSYN with the configure modified.

• Adaptive Sampling Strategy
• Adaptive Threshold System
• Data Quality Enhanced Implementation
• Adaptive Reward System: Dynamic quality enhanced and dataset weighting for optimal learning
• Multi-domain Coding: Trained on MBPP and CodeContests datasets with adaptive sampling

The model is trained on a composite dataset with adaptive weighted sampling strategy

For detail on Huggingface → Qwen2.5-Coder-0.5B-Instruct-Gensyn-Swarm-My-Agent-ID

• Qwen/Qwen2.5-Coder-0.5B-Instruct → Recommend Solver
• deepseek-ai/deepseek-coder-1.3b-instruct → Proposers, Evaluators
• Qwen/Qwen2.5-Coder-1.5B-Instruct → Proposers, Evaluators

I. Update System Packages

apt update && apt upgrade -y && \
apt install screen curl ufw nload tree iptables git wget lz4 jq make gcc nano automake autoconf \
htop tmux libgbm1 protobuf-compiler python3 python3-pip python3-venv python3-dev python3-setuptools \
tar clang nethogs ncdu unzip build-essential pkg-config libssl-dev libleveldb-dev \
speedtest-cli ca-certificates libffi-dev libsqlite3-dev -y

II. Install v22 Node.js - Npm - Yarn - Pm2

source <(wget -qO- https://raw.githubusercontent.com/arcxteam/w-ai-wombo/main/nodejs.sh)

III. Clone Repository

git clone https://github.com/arcxteam/rl-swarm.git && cd rl-swarm

I. If run with OctaSpace GPU default Tmux/Termux sessions

• Full comprehensif guide tips & trick read here Octaspace
• Quick-1: Always rent with this template UBUNTU 22.04/24.04 LTS dont use another template
• Quick-2: Always deploy in coloum (Expose HTTP Ports) input https port as 3000 will auto create localhost to access gensyn login
• Create new sessions CTRL+B+C
• Go to sessions tmux attach
• List sessions CTRL+B+W
• Navigate sessions scroll up ↑ down ↓ or click any sessions & then enter
• Close with run background CTRL+B+D

II. If run with other cloud GPU use Screen sessions

• Create sessions screen -S gensyn
• Close sessions CTRL+A+D
• Go to sessions screen -r gensyn

III. Virtual Python & Running

python3 -m venv .venv

source .venv/bin/activate
# if not worked, then:
. .venv/bin/activate

./run_rl_swarm.sh

I. Install Docker & Compose → if the cloud GPU was support & use Ubuntu VM

curl -sSL https://raw.githubusercontent.com/arcxteam/succinct-prover/refs/heads/main/docker.sh | bash

II. Setup build on docker

• CPU

docker compose run --rm --build -Pit swarm-cpu

• GPU

docker compose run --rm --build -Pit swarm-gpu

I. Get API key: https://wandb.ai/authorize II. Create new Project and edit this or nano .env

# Create .env file
cat > .env << 'EOF'
WANDB_MODE=online
WANDB_API_KEY=YOUR_API_KEY
WANDB_ENTITY=YOUR_TEAM_OR_USERNAME
WANDB_PROJECT=YOUR_PROJECT_NAME
EOF

III. Execute script

chmod +x wandb_run_rl_swarm.sh

IV. Running*

wandb online
./wandb_run_rl_swarm.sh

• Open a new terminal
• Install localtunnel

npm install -g localtunnel

• The password is actually your VPS-IP

curl ifconfig.me && echo

• Get URL access

lt --port 3000

Visit the prompted url, and enter your password to access Gensyn login page

All GGUF formats are llama.cpp is compatible ready to use Inferences chat and auto-update be hourly.

• Gensyn Documentation: https://docs.gensyn.ai/
• Gensyn GitHub: https://github.com/gensyn-ai
• RL-Swarm Contracts: https://github.com/gensyn-ai/rl-swarm-contracts
• Qwen2.5-Coder Model Card: https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B-Instruct
• MBPP Dataset: https://huggingface.co/datasets/google-research-datasets/mbpp
• CodeContests Dataset: https://huggingface.co/datasets/deepmind/code_contests
• arXiv:1910.09700: ML Carbon Emissions methodology
• Community: Gensyn Discord