OPTIMIZATION & SWARMS IN DEVELOPMENT LOCAL AIR-GAPPED
AutoMLApp Node-Graph Guide
Visual pipeline editor with 185+ node types across 25+ domains, F7 invariant validation, and HyperSwarm integration.
AutoMLApp Node-Graph Guide
AutoMLApp is a high-performance Windows desktop node-graph editor (.NET 10 / WinForms) for designing, validating, and executing machine learning, hardware accelerator, and swarm optimization pipelines.
Key Features
- 185+ Node Types across 25+ Domains: Data pipelines, ONNX models, IPU tile configurations, spiking networks, swarm optimizers, FPGA bitstreams, and telemetry sidecars.
- Instant F7 Invariant Validation: Runs static graph analysis to flag cycles, disconnected ports, orphaned nodes, and port type mismatches before execution.
- Topological Plan Compilation: Compiles the visual canvas into standardized execution plans dispatchable to local accelerators or remote nodes.
- HyperSwarm-Backed AutoML: Automated neural architecture and hyperparameter search with cost-aware telemetry scoring.
Launching & Invocations
1. Launch Visual Desktop Canvas
dotnet run --project src/HFLabs/Products/AutoMLApp/AutoMLApp.csproj -c Release2. Headless SwarmML Optimizer
dotnet run --project src/HFLabs/Products/AutoMLApp/SwarmML.csproj -- optimize --iterations 500 --config pipeline.jsonNode Connector Types & Validation Rules
Ports enforce strict type compatibility. Canvas links are rejected in real-time if schemas mismatch:
Tensor<T>: Multi-dimensional array tensors (FP32, FP16, BF16, INT8).GraphContext: In-memory property graph and embedding states.SwarmConfig: Population, velocity limits, and fitness objective handlers.IPUTileMap: Spatial tile allocations for Graphcore Colossus processors.