2 projects
deer-probe
DEER is an encoder-based knowledge graph completion (KGC) model that uses embedding vectors from generative language models for few-shot learning. It retains in-context learning while ensuring efficient large-scale inference without fine-tuning. DEER excels at predicting new relation types in small KGs and aligns with LAMA for knowledge probing, making it a scalable tool for evaluating factual knowledge in PLMs.
