Matches in Nanopublications for { <https://neverblink.eu/ontologies/llm-kg/methods#KGo1> ?p ?o ?g. }
Showing items 1 to 10 of
10
with 100 items per page.
- KGo1 type Workflow assertion.
- KGo1 type Workflow assertion.
- KGo1 label "KG-o1" assertion.
- KGo1 label "KG-o1" assertion.
- KGo1 comment "KG-o1 is a four-stage framework designed to enhance the intrinsic multi-hop reasoning abilities of LLMs. It leverages KGs to filter entities, generate logical paths, construct complex QA datasets for supervised fine-tuning (SFT) to simulate long-term thinking, and applies a Self-improved Adaptive DPO strategy to refine the LLMs' reasoning, ultimately improving LLM performance during inference for multi-hop question answering." assertion.
- KGo1 comment "KG-o1 is a novel four-stage framework that integrates Knowledge Graphs (KGs) to enhance Large Language Models' (LLMs) multi-hop reasoning abilities. It involves constructing KG-derived datasets (KG-MHQA SFT and DPO) and using them to fine-tune LLMs (via Supervised Fine-Tuning and a "Self-improved Adaptive DPO" strategy), aiming to improve the LLM's intrinsic knowledge expression and reasoning capabilities during a training stage by internalizing logical paths." assertion.
- KGo1 subject KGEnhancedLLMInference assertion.
- KGo1 subject KGEnhancedLLMPretraining assertion.
- KGo1 hasTopCategory KGEnhancedLLM assertion.
- KGo1 hasTopCategory KGEnhancedLLM assertion.