[{"_1":2,"_168":-5,"_169":-5},"loaderData",{"_3":4,"_13":14,"_21":22,"_145":146},"root",{"_5":-5,"_6":-7,"_7":8,"_9":-7,"_10":11,"_12":-7},"currentUser","primaryEmail","siteVerification","6LeIxAcTAAAAAJcZVRqyHh71UMIEGNQ_MXjiZKhI","toast","enableDev",false,"incidentStatus","routes/_layout",{"_15":-5,"_16":17,"_18":17,"_19":11,"_20":11},"username","isAskGordonEnabled",true,"isUsingNewTemplate","showMinimalLanding","showHoneypotTrap","routes/_layout.r.$namespace.$reponame",{"_23":24,"_25":-5,"_26":11,"_27":11,"_28":29,"_30":31,"_32":-5,"_33":11,"_34":35,"_88":-5,"_89":90,"_136":-5,"_137":11,"_138":17,"_139":11,"_140":-5,"_141":11,"_142":-5,"_143":144},"canonicalUrl","https://hub.docker.com/r/djstomp/shadowswarm-app/tags","mcpServer","canSeeDockerfile","canSeeAutobuilds","namespace","djstomp","reponame","shadowswarm-app","cloudBuildSource","isAutobuildsAvailable","repo",{"_36":37,"_43":44,"_46":47,"_48":49,"_50":11,"_51":52,"_53":54,"_55":56,"_39":31,"_28":29,"_58":59,"_60":45,"_61":62,"_63":64,"_65":66,"_67":-5,"_68":62,"_69":70,"_71":11,"_72":29,"_73":74,"_79":11,"_80":81,"_85":29,"_86":87},"categories",[38],{"_39":40,"_41":42},"name","Machine learning & AI","slug","machine-learning-and-ai","content_types",[45],"image","date_registered","2025-01-14T10:55:30.413221Z","description","A streamlined framework for distribution of GPU-accelerated PyTorch workloads via Docker Swarm","is_private","last_updated","2025-01-14T13:22:49.623565Z","last_modified","2025-01-14T11:30:26.989938Z","media_types",[57],"application/vnd.oci.image.index.v1+json","pull_count",934,"repository_type","star_count",0,"status",1,"status_description","active","source","collaborator_count","full_description","

ShadowSwarm

\n\nA streamlined framework for setting up a multi-node, GPU-accelerated, distributed system for PyTorch workloads using Docker Swarm. With **ShadowSWARM**, you can quickly configure and deploy a scalable environment for machine learning inference or training across multiple machines.\n\n

\n\n## **Features**\n- Automated Docker Swarm initialization and worker node setup.\n- Flexible configuration using interactive CLI (`config.py`).\n- Dynamic IP and hostname detection for seamless multi-node deployment.\n- Streamlined distributed PyTorch workloads with Fully Sharded Data Parallel (FSDP).\n- Integrated **Streamlit** interface for easy interaction with your system.\n\n## **Quickstart Guide**\n\n### **Prerequisites**\n\n1. **Docker and NVIDIA Drivers**:\n - Install Docker and NVIDIA drivers on all machines.\n - Install the NVIDIA Container Toolkit:\n ```bash\n sudo apt-get install -y nvidia-container-toolkit\n sudo systemctl restart docker\n ```\n - Verify Docker GPU support:\n ```bash\n docker run --rm --gpus all nvidia/cuda:12.1.1-base-ubuntu20.04 nvidia-smi\n ```\n\n2. **Python 3.8+**:\n - Install Python on the master machine:\n ```bash\n sudo apt-get install python3 python3-pip\n ```\n\n3. **Passwordless SSH**:\n - Configure **passwordless SSH from the master to all worker nodes**:\n ```bash\n ssh-keygen -t rsa -b 2048\n ssh-copy-id user@worker-ip\n ```\n - You only need to set up SSH from the **master node to the workers**.\n - The worker nodes do **not** need SSH access to each other or the master.\n\n### **Installation**\n\n1. **Clone the Repository (Only on Master)**:\n - Clone this repository on the master node:\n ```bash\n git clone https://github.com/DJStompZone/shadowswarm.git\n cd shadowswarm\n ```\n - The worker nodes do **not** need the repository because Docker Swarm handles the deployment of containers automatically.\n\n2. **Build the Docker Image**:\n Build the Docker image on the master node:\n ```bash\n docker build -t shadowswarm-app .\n ```\n\n### **Setup and Deployment**\n\n1. **Run the Configuration Script**:\n Use the interactive CLI to gather and validate the necessary configuration:\n ```bash\n python3 config.py\n ```\n This script will:\n - Prompt for the master and worker node details.\n - Save the configuration to a `.env` file.\n - Start the `bootstrap.sh` script to initialize Docker Swarm and add workers.\n\n2. **Verify Swarm Setup**:\n Check the Swarm status after the bootstrap:\n ```bash\n docker node ls\n ```\n\n3. **Deploy the Docker Stack**:\n Once the Swarm is ready, deploy the application:\n ```bash\n docker stack deploy --compose-file docker-compose.yml shadowswarm\n ```\n\n\n### **Access the Streamlit App**\n\n1. Open a browser and navigate to the **master node IP**:\n ```\n http://:8501\n ```\n\n2. Use the **Streamlit interface** to interact with your distributed PyTorch system.\n\n\n## **File Structure**\n\n```\nshadowswarm/\n├── config.py # CLI script for gathering configuration\n├── bootstrap.sh # Script for initializing Docker Swarm and adding workers\n├── docker-compose.yml # Docker Swarm stack configuration\n├── Dockerfile # Docker image definition\n├── .env # Environment variables for the deployment\n├── app/ # Application directory\n│ ├── main.py # PyTorch and Streamlit code\n│ └── utils.py # Utility functions\n```\n\n\n## **How It Works**\n\n1. **Configuration**:\n - `config.py` prompts for master and worker node details, saves them to `.env`, and triggers `bootstrap.sh`.\n\n2. **Swarm Initialization**:\n - `bootstrap.sh` initializes Docker Swarm on the master node and connects workers via SSH.\n\n3. **Stack Deployment**:\n - `docker-compose.yml` orchestrates the master and worker containers, assigning roles using environment variables.\n\n4. **Distributed Workload**:\n - The master node manages the distributed PyTorch workload across all nodes using Fully Sharded Data Parallel (FSDP).\n\n\n## **Environment Variables**\n\n| Variable | Description |\n|--------------------|--------------------------------------------|\n| `MASTER_HOSTNAME` | Hostname of the master node. |\n| `MASTER_IP` | IP address of the master node. |\n| `WORKER_HOSTNAMES` | Comma-separated list of worker hostnames. |\n| `NODE_RANK` | Rank of the node in the distributed setup. |\n| `WORLD_SIZE` | Total number of nodes in the cluster. |\n| `MASTER_PORT` | Port for master-worker communication. |\n\n## **Troubleshooting**\n\n1. **Docker Swarm Issues**:\n - Check if Swarm is initialized:\n ```bash\n docker info\n ```\n - Verify worker nodes are connected:\n ```bash\n docker node ls\n ```\n\n2. **SSH Issues**:\n - Test passwordless SSH from the master:\n ```bash\n ssh \n ```\n\n3. **Container Logs**:\n - Check the logs for the master or workers:\n ```bash\n docker service logs shadowswarm_master\n docker service logs shadowswarm_worker1\n ```\n\n4. **GPU Issues**:\n - Ensure GPUs are accessible:\n ```bash\n docker run --rm --gpus all nvidia/cuda:12.1.1-base-ubuntu20.04 nvidia-smi\n ```\n \n## **Scaling**\n\n1. Add a new worker node to the swarm:\n ```bash\n docker swarm join --token

:2377\n ```\n\n2. Update the `WORKER_HOSTNAMES` in the `.env` file to include the new worker.\n\n3. Re-deploy the stack:\n ```bash\n docker stack deploy --compose-file docker-compose.yml shadowswarm\n ```\n\n## **Contributing**\n\nContributions are welcome! Please open an issue or submit a pull request if you have problems, suggestions, or improvements.\n\n## **License**\n\nThis project is licensed under the [MIT License](LICENSE).","has_starred","hub_user","immutable_tags_settings",{"_75":11,"_76":77},"enabled","rules",[78],".*","is_automated","permissions",{"_82":17,"_83":11,"_84":11},"read","write","admin","user","storage_size",11122226782,"orgData","recentTags",[91],{"_92":45,"_93":94,"_95":96,"_97":98,"_99":100,"_101":102,"_51":125,"_126":94,"_127":29,"_128":57,"_39":129,"_130":131,"_132":110,"_133":125,"_134":66,"_135":17},"content_type","creator",13690372,"digest","sha256:e8098d2bfb8d8e129629f824d580cbdcf6cb8a7768d6a459b3525d0db557fe37","full_size",4682301378,"id",829115443,"images",[103,119],{"_104":105,"_95":106,"_107":108,"_109":110,"_111":112,"_113":114,"_115":108,"_116":-5,"_117":98,"_63":66,"_118":-5},"architecture","amd64","sha256:d93dd6f068a923447fb72993edbcd73e15bc38708e21f0a719e4c56a2e49de76","features","","last_pulled","2026-09-22T08:20:32.993397317Z","last_pushed","2025-01-14T12:39:20.701894973Z","os","linux","os_features","os_version","size","variant",{"_104":120,"_95":121,"_107":108,"_109":122,"_111":123,"_113":120,"_115":108,"_116":-5,"_117":124,"_63":66,"_118":-5},"unknown","sha256:7bcf094addf225946a3e7ee53e303b06a13fec9d9fe1f8c6a267f01644955c0f","2026-08-30T19:21:51.124805192Z","2025-01-14T13:13:43.481409154Z",1193,"2025-01-14T13:22:48.794019Z","last_updater","last_updater_username","media_type","latest","repository",26186356,"tag_last_pulled","tag_last_pushed","tag_status","v2","sbxKitMetadata","isPotentialSpam","isHubSbxKitsEnabled","showSbxLogoOverride","sbxTrademarkText","displaySecurityMetadata","tagAttestations","tagsHaveAttestations",[],"routes/_layout.r.$namespace.$reponame.tags",{"_34":147,"_155":156,"_143":166,"_167":11},{"_36":148,"_43":150,"_46":47,"_48":49,"_50":11,"_51":52,"_53":54,"_55":151,"_39":31,"_28":29,"_58":59,"_60":45,"_61":62,"_63":64,"_65":66,"_67":-5,"_68":62,"_69":70,"_71":11,"_72":29,"_73":152,"_79":11,"_80":154,"_85":29,"_86":87},[149],{"_39":40,"_41":42},[45],[57],{"_75":11,"_76":153},[78],{"_82":17,"_83":11,"_84":11},"tags",{"_157":64,"_158":-5,"_159":-5,"_160":161},"count","next","previous","results",[162],{"_92":45,"_93":94,"_95":96,"_97":98,"_99":100,"_101":163,"_51":125,"_126":94,"_127":29,"_128":57,"_39":129,"_130":131,"_132":110,"_133":125,"_134":66,"_135":17},[164,165],{"_104":105,"_95":106,"_107":108,"_109":110,"_111":112,"_113":114,"_115":108,"_116":-5,"_117":98,"_63":66,"_118":-5},{"_104":120,"_95":121,"_107":108,"_109":122,"_111":123,"_113":120,"_115":108,"_116":-5,"_117":124,"_63":66,"_118":-5},[],"canDeleteTags","actionData","errors"]

djstomp/shadowswarm-app

Tags for djstomp/shadowswarm-app