Hillock: Non-Generative CUDA Pipeline for Fast Knowledge Graph Extraction

Equivalent-Flan-1590 · reddit · 2026-08-12

Developer roandejager open-sources Hillock, whose v0.2 introduces TALON, a non-generative tensor pipeline for KG triple extraction. It uses Fastcoref, MiniLM-L6-v2 routing, and GLiREL classification. Processes 32 sentences in 2s on GTX 1070 with <1GB VRAM, 50% recall. AGPL-3.0.

Related event: Open-Sourcing Hillock: Accelerating Agent Memory with Non-Generative Tensor Pipelines(2 posts)→

Original post →

More from coding & agent

coding & agent channel →