Knowledgator Model Hub
Knowledgator provides a collection of high-performance models tailored for a wide range of information extraction tasks. Our models are built to support real-world applications in both zero-shot and few-shot settings, enabling efficient deployment even with minimal labeled data.
Model Categories
Multi-task
Models capable of handling multiple NLP tasks simultaneously — ideal for pipelines requiring shared representations across domains.
Text-to-JSON
Extracting structured JSON of any complexity from a text, given a user-provided template.
NER (Named Entity Recognition)
Extract domain-specific entities with high precision using models fine-tuned for biomedical, legal, and general-purpose use cases.
Text Classification
Classify text into topics, sentiments, or intents. Includes:
- ComprehendIt – A general-purpose, zero-shot NLI-based classifier.
- GLiClass – A lightweight, fast model for resource-constrained environments.
Entity Linking
Bi-encoder models for entity disambiguation that link extracted mentions to knowledge base entries. Includes:
- GLiNER-Linker Base/Large -- DeBERTa-based linking models for entity disambiguation.
- GLiNER-Linker Rerank -- Lightweight reranker for improved accuracy on large candidate sets.
Chemical Models
Specialized models for chemical domain tasks such as converting between SMILES and IUPAC names.
Features
- Zero-shot ready: pretrained for strong performance without task-specific training
- Few-shot efficient: achieves superior results with minimal labeled examples
- Multi-task support: handles multiple information extraction tasks in a unified framework
- Hugging Face compatible: works seamlessly with the Hugging Face Transformers ecosystem
- Run locally: supports fully local inference for privacy, security, and offline use
- Domain adaptable: effective across diverse domains, including general, biomedicine, legal, finance, and chemistry
Knowledgator – powering accurate and adaptable information extraction at scale.