knowgl-large
| Property | Value |
|---|---|
| Parameter Count | 406M |
| License | CC-BY-NC-SA-4.0 |
| Architecture | BART-based Transformer |
| Paper | KnowGL Paper |
| Performance | 70.74% RE+ Macro F1 |
What is knowgl-large?
knowgl-large is a state-of-the-art model designed for knowledge generation and linking from text. Developed by IBM, it combines Wikidata with an extended version of the REBEL dataset to generate structured knowledge triples from natural language text. The model employs a sophisticated approach to extract relationships between entities while maintaining their semantic context.
Implementation Details
Built on the BART architecture, this 406M parameter model generates triples in a specialized format: [(subject mention # subject label # subject type) | relation label | (object mention # object label # object type)]. Multiple triples are separated by '$' in the output, making it ideal for complex knowledge extraction tasks.
- Trained on combined Wikidata and REBEL dataset
- Generates typed entity and relation triples
- Direct mapping capability to Wikidata IDs
- F32 tensor type implementation
Core Capabilities
- Relation Extraction with SOTA performance
- Entity Linking and Type Recognition
- Triple Generation for Knowledge Graphs
- Semantic Relationship Mapping
Frequently Asked Questions
Q: What makes this model unique?
The model's ability to generate complete knowledge triples with entity types and labels, while maintaining direct mappability to Wikidata IDs, sets it apart from traditional relation extraction models.
Q: What are the recommended use cases?
The model is ideal for knowledge graph construction, information extraction from text, and automated relationship mapping in corporate research and data analysis scenarios.
