HFLabs Machine Learning Suite
MACHINE LEARNING SUITE LOCAL ML STACK LOCAL AIR-GAPPED

HFLabs Machine Learning Suite

Hyperparameter optimization, knowledge graph embeddings, local Level Zero LLM inference, and SNN simulation in .NET.

#ML Suite#HPO#KGE#Level Zero LLM#SNN

HFLabs Machine Learning Suite

The HFLabs Machine Learning Suite provides composable, dependency-free .NET libraries for model tuning, knowledge graph representation learning, local LLM execution, and neuromorphic spiking networks—with zero cloud services in the loop.

Core Modules

1. Hyperparameter Optimization (HPO)

A resumable optimization framework built around disk-checkpointed sessions:

  • 5 Search Algorithms: Bayesian Optimization (Gaussian Process surrogate), Hyperband, BOHB, ASHA-BOHB, and TuRBO (Trust-Region Bayesian Optimization).
  • JSON Checkpoints: Searches survive power loss or system restarts seamlessly.
  • Continuous, Integer & Categorical parameters with early-stopping criteria.

2. Knowledge Graph Embeddings (KGE)

  • 6 Scoring Functions: TransE, RotatE, ComplEx, DistMult, ZPinch-TransE (anchors schema roots near origin), and Dual-Conjugate ComplEx (inverse relations).
  • Dual Hardware Acceleration: SIMD CPU TensorPrimitives + Poplar-native IPU engine.
  • Auto-Extractors: Generate graphs from codebases, database schemas, architectural plans, skill files, and white papers.

3. Local LLM Inference (Level Zero)

  • From-scratch transformer runtime in pure C# / .NET 10.
  • Native Hugging Face .safetensors loader, BPE tokenizer, and Qwen chat templating.
  • Full decoder layers, RoPE embeddings, linear attention, and dynamic KV-cache.
  • Direct Intel Level Zero GPU offload with companion l0llm CLI tool.

4. Graph-Constrained Decoding

  • Tri-beam decoder synchronizing beam candidates against knowledge graph entity boundaries.
  • Token trie constructed over graph entities prevents hallucinations by construction during token generation.

5. Spiking Neural Networks (SNN)

  • Reference CPU simulator for neuromorphic experiments.
  • Leaky Integrate-and-Fire (LIF) neuron groups, event-driven spike scheduling, refractory periods, and Spike-Timing-Dependent Plasticity (STDP).