Skip to main content

Knowledgator Datasets Hub

Knowledgator offers a curated collection of high-quality datasets tailored for training and evaluating models across a wide range of information extraction tasks. These datasets are optimized for real-world scenarios, support multi-lingual and domain-specific use cases, and are compatible with Knowledgator frameworks like GLiNER, GLiClass, and LiqFit.

Dataset Categories​

General Purpose​

Versatile datasets designed for broad NLP tasks such as text classification, relation extraction, and entity recognition.

  • knowledgator/GLINER-multi-task-synthetic-data: Synthetic multi-task dataset for NER, RE, summarization, sentiment, QA, and more.
  • gliclass-v2.0 / gliclass-v2.0-RAC: Text classification dataset with optional retrieval-augmented context.

Multilingual​

Cross-lingual datasets supporting named entity recognition in 10+ languages.

  • tner/multinerd: Multilingual, multi-genre NER annotations.
  • MultiCoNER/multiconer_v2: Handles low-context and complex named entities across 12 languages.
  • unimelb-nlp/wikiann: WikiANN dataset for 176 languages.

Biomedical​

Specialized corpora for biomedical and chemical domain tasks such as disease recognition, gene tagging, and adverse effect extraction.

  • bigbio/anat_em, drugprot, chia, nlm_gene, scai_disease, and more: Cover anatomical, pharmaceutical, and clinical trial data.

Highlights​

  • Supports NER, RE, Text Classification, Keyphrase Extraction, QA, and more.
  • Datasets annotated manually, synthetically, or with weak supervision.
  • Available via Hugging Face Datasets or directly through Knowledgator tools.
  • Structured to work with few-shot and retrieval-augmented learning.