Getting Started with HFLabs
OVERVIEW & SETUP QUICKSTART LOCAL AIR-GAPPED

Getting Started with HFLabs

Quickstart guide, prerequisites, repository layout, and verification steps for HFLabs.

#Setup#.NET 10#Level Zero#IPU

Getting Started with HFLabs

Welcome to the HFLabs developer documentation. This guide walks you through system prerequisites, repository structure, and verified command recipes to build, test, and run HFLabs products on your local workstation.

Core Design Principles

HFLabs systems are designed from first principles around three strict architectural rules:

  1. Zero Cloud Dependencies in the Loop: All optimization, knowledge graph inference, and LLM execution runs 100% locally on your workstation or air-gapped on-premises cluster.
  2. Deterministic & Inspectable Execution: Pipelines emit structured telemetry sidecars and detailed audit trails at every step.
  3. Hardware Agnostic with Accelerators: CPU SIMD (AVX-512 / Neon) provides the baseline; Intel Level Zero GPUs and Graphcore IPUs act as optional high-throughput accelerators.

Prerequisites

Ensure your environment has the following toolchains installed:

  • .NET 10 SDK (or .NET 9.0+) with C# 13 language support.
  • PowerShell 7+ (pwsh) for cross-platform automation scripts.
  • Intel OneAPI Level Zero Loader & Drivers (optional, required for Intel Iris Xe / Arc GPU acceleration).
  • Graphcore Poplar SDK (optional, required for IPU compute).

Repository Organization

src/HFLabs/
├── Products/
│   ├── HypertensorGraphRAG/      # In-memory property graph & vector engine
│   ├── HyperSwarmOptimization/  # 35 swarm optimizers & Director ensemble
│   ├── SwarmDev/                 # 6-stage deterministic pipeline
│   ├── AutoMLApp/                # WinForms node-graph editor (.NET 10)
│   └── QueryAccel/               # IPU SQL query acceleration extension
├── Libraries/
│   └── ML/                       # HPO, KGE, Level Zero LLM & SNN engines
├── Hardware/
│   ├── LevelZero.NET/            # Managed Intel Level Zero GPU compute wrapper
│   ├── KernelCatalog/            # 26 precompiled SPIR-V & Zebin OpenCL C kernels
│   ├── Accelerator/              # gRPC PlanEnvelope heterogeneous orchestrator
│   └── Poplar.NET/               # Graphcore Poplar/PopART driver bindings
└── DevTools/
    └── ABIGen/                   # libclang C/C++ header binding synthesizer

Quick Verification Recipe

Run the baseline smoke tests to ensure your local environment is configured:

# 1. Run GraphRAG headless test query
dotnet run --project src/HFLabs/Products/HypertensorGraphRAG/App/App.csproj -c Release -- --headless --query "Director"

# 2. Run HyperSwarm benchmark sweep
dotnet run --project src/HFLabs/Products/HyperSwarmOptimizationSuite/BenchmarkRunner.csproj -c Release -- matrix-sweep

# 3. Audit Level Zero precompiled SPIR-V kernels
pwsh -File src/HFLabs/Hardware/LevelZero.Toolchain/Audit-SpvKernels.ps1