zjunlp/DeepKE

Supports entity, relation, attribute, and event extraction across specialized scenarios including few-shot learning, document-level processing, multimodal inputs, and Chinese schema-based tasks. Implements state-of-the-art architectures like W2NER, PRGC, and ASP alongside LLM-based extraction via instruction tuning and in-context learning. Integrates with Hugging Face Transformers and provides pre-trained models, MCP tool integration, and comprehensive datasets (IEPile, InstructIE) for supervised and instruction-based knowledge graph construction.

4,338 stars and 751 monthly downloads. No commits in the last 6 months. Available on PyPI.