relik

Why we included this project

ReLiK handles a fiddly slice of text understanding: finding the named entities in a document, mapping each one to a canonical entry in a knowledge base, and extracting the relations that connect them. Its retrieve-read-link design keeps latency and memory low enough to run on a single consumer GPU, a practical concern for teams without large clusters. The repo ships pretrained retriever and reader models on Hugging Face, a CLI, serving endpoints, and a training pipeline, so you can jump from the demo to fine-tuning on your own data without wiring separate tools together. Researchers will also find the ACL 2024 paper and reference implementation a solid starting point for entity linking and relation extraction work.