Knowledgator Engineering launched GLiFormer, a schema-conditioned framework for information extraction.
Two checkpoints are available: Base v1 has 264.2M parameters, and Large v1 has 575.6M. Both models use Apache 2.0 licenses and install via pip on CPU or GPU.
The system encodes source text once to handle multiple extraction tasks simultaneously. It supports named-entity recognition, classification, relation extraction, and nested JSON structuring. Values come directly from the input text, preventing hallucinated content.
