DKnown-Jev Classifies Normative Knowledge Usability in 2B and 27B Models
October 20, 2026
Fine-tuned Qwen3.5-2B and Qwen3.8-27B models classify text as Answer Evidence (AE) or Non-answer Evidence (NE) for normative tasks like eligibility determination. The models use metadata and question conditions to assess if a paragraph satisfies specific rules and requirements.
HOW THIS AFFECTS YOU
●
builderYou can use these configurations to improve accuracy in RAG pipelines involving legal or regulatory documentation.
●
researcherYou can utilize these fine-tuned weights for domain-specific rule compliance tasks.