How hewlettpackard/swarm-learning works
Swarm Learning is a decentralized framework for training models across data-owning sites while exchanging learned weights rather than raw data. The principal workflow is to configure and deploy a run, train locally through a platform-specific callback, synchronize model weights with the local Swarm Learning node, and repeat until training completes; SWCI and SLM-UI support management and monitoring. The Python client and orchestration relationships below are supported by excerpts or the README. Other runtime components are documented capabilities represented at their tree-supported directories; their internal implementation was not sampled.
Main entry points: swarm.py Swarm callback · Working_of_a_Swarm_Learning_node.md SL node · Swarm_Learning_architecture.md Swarm Network · Swarm_Operator_node_(SWOP).md SWOP tasks
Components 17
Deployment and management
- SWOP tasks
docs/User/Swarm_Operator_node_(SWOP).md - SLM-UI
docs/User/Monitoring_Swarm_Learning_training_using_SLM-UI.md - AutoPass License Server
docs/HPE AutoPass License Server User Guide.pdf
Model training
- User ML program
examples - TensorFlow and Keras
lib/src/python-client/swarmlearning/tf.py - PyTorch
lib/src/python-client/swarmlearning/pyt.py - Scikit-learn
lib/src/python-client/swarmlearning/sklearn.py - Hugging Face
lib/src/python-client/swarmlearning/hf_transformers.py - Swarm callback
lib/src/python-client/swarmlearning/client/swarm.py
Swarm coordination
- Application interface
lib/src/python-client/swarmlearning/client/app2if.py - IPC bridge
lib/src/python-client/swarmlearning/com/bridge.py - SL node
docs/User/Working_of_a_Swarm_Learning_node.md - Swarm Network
docs/User/Swarm_Learning_architecture.md - Blockchain metadata
docs/User/Swarm_Learning_architecture.md
Monitoring
- SWCI
lib/src/python-client/swarmlearning/swci.py - Training progress
docs/User/Monitoring_Swarm_Learning_training_using_SLM-UI.md
Other components
- Operator
Connections 18
- Operator → SWOP tasks: assigns tasks
README.md - Operator → SLM-UI: deploys run
README.md - SWOP tasks → User ML program: launches training
README.md - SLM-UI → Training progress: monitors runs
README.md - Operator → SWCI: manages network
README.md - SWCI → Swarm Network: connects to
README.md - User ML program → Swarm callback: uses callback
README.md - TensorFlow and Keras → Swarm callback: extends
lib/src/python-client/swarmlearning/tf.py - PyTorch → Swarm callback: extends
lib/src/python-client/swarmlearning/pyt.py - Scikit-learn → Swarm callback: uses base
lib/src/python-client/swarmlearning/sklearn.py - Hugging Face → Swarm callback: extends
lib/src/python-client/swarmlearning/hf_transformers.py - Swarm callback → Application interface: invokes interface
lib/src/python-client/swarmlearning/client/swarm.py - Application interface → IPC bridge: sends messages
lib/src/python-client/swarmlearning/client/app2if.py - IPC bridge → SL node: exchanges weights
README.md - SL node → Swarm Network: coordinates through
README.md - Swarm Network → Blockchain metadata: tracks metadata
README.md - SL node → Training progress: reports progress
README.md - Operator → AutoPass License Server: licenses framework
README.md