SLKR: big LLM builds label knowledge offline, small LLM classifies zero-shot via retrieval
_reachsumit · x · 2026-10-14
A paper proposes SLKR (Semantic Label Knowledge Retrieval), a label-centric RAG framework for zero-shot text classification that separates offline knowledge construction from lightweight online inference.
- Motivation: flagship LLMs are too costly and slow for large-scale labeling; the goal is zero-shot classification without labeled examples or classifier feedback
- Offline: a flagship LLM generates label definitions, retrieval keywords, and boundary rules for confusable label pairs into a reusable label-knowledge store; construction is over the label space, so the one-time cost amortizes across large-scale inference
- Online: hybrid semantic + lexical retrieval picks top-K candidate labels and up to M input-relevant boundary rules; a lightweight LLM classifies in a single call
- Results: across three public benchmarks and a real-world multilingual book-classification dataset, SLKR wins on all four datasets with Qwen3.5-4B and three of four with Ministral-3-3B, at much lower compute cost (abstract truncated)
More from Research
- FrontierML benchmark tests whether agents can spend money to scale model performance — JayAlammar · 2026-10-14
- AIPOCH's Open-Science ships portable .science packages that share full research context, not just outputs — ahuja_priyank · 2026-10-14
- Video-Arena: Game-Based Video Benchmark Shows Native Models Score Just 13.2 F1 — Kuan Zhang · 2026-10-14
- LSCBench: Vibe Coding Agents Write Functional Code, But Only 25% Is Also Secure — Qwen · 2026-10-14
- Omni-LLMs Know More Than They Say: NAVER's OmniCatalyst Adds 7.29% via Self-Generated Text — naver-ai · 2026-10-14
- Repeated-Data Pretraining Study: Smaller Batches Act as Regularizers and Beat More GPUs — Sewon Min · 2026-10-14