Skip to main content

Knowledgator Frameworks

Knowledgator offers a suite of modular frameworks designed to accelerate the development of information extraction systems across diverse domains. These tools provide lightweight, task-specific solutions and support seamless integration with Hugging Face Transformers, making them easy to adopt for both research and production use.

Available Frameworks​

GLiNER​

A token classification framework optimized for Named Entity Recognition (NER).

  • Effective across text classification, QA, NER, Relation Extraction.
  • Supports few-shot and zero-shot entity extraction
  • Works with custom and overlapping entity types
  • Multiple architecture types

GLiClass​

A simplified framework for text classification, focused on minimal setup and high efficiency.

  • Supports few-shot and zero-shot text classification
  • Fast inference
  • Multi-label and multi-class support
  • Multiple architecture types

LiqFit​

A flexible framework for few-shot learning of cross-encoder models using natural language inference (NLI).

  • Effective across text classification, QA, NER, and more
  • Minimal data requirements (as few as 8 examples per label)
  • Built-in support for DeBERTa, T5, and custom heads

Key Benefits​

  • Task-specific abstractions with minimal boilerplate
  • Support for fine-tuning, inference, and evaluation
  • Open-source and compatible with Hugging Face ecosystem
  • Designed for robust generalization across domains