AtlasKV: billion-scale knowledge graphs in 20GB VRAM, no retriever or vector DB needed
udmrzn · x · 2026-11-20
Researchers from HKUST and Huawei's Theory Lab propose AtlasKV, a parametric knowledge augmentation method that injects billion-scale knowledge graph triples (e.g. 1B) directly into an LLM's inherent attention mechanism using under 20GB VRAM. Via two new methods (KG2KV and HiKVP) it achieves sub-linear time and memory complexity, with no external retrievers, no long-context priors, and no retraining when adding new knowledge — a potential replacement for the whole RAG pipeline. Code and data are open-sourced.
More from Infra
- Wedbush raises Nvidia price target to $293, says leadership risks already priced in — sudoraohacker · 2026-11-20
- DeepMind essay maps four paths to superintelligence: effective compute grows 10x per year — udmrzn · 2026-11-20
- 2M canary-token prompts reveal AI provider breaches and prompt theft — WeldPond · 2026-11-20
- Open weights jump from 7.3% to 30.9% of Databricks-hosted tokens in four months — AIFlow_ML · 2026-11-20
- 7 web search APIs with MCP servers for Claude and Cursor, benchmarked — JeremyCMorgan · 2026-11-20
- Baseten ships deferred tool loading: agents run up to 62% cheaper and 26% faster — baseten · 2026-11-20