Knowledgator unveils GLiFormer: a high-performance encoder for nested json extraction

Knowledgator has introduced GLiFormer, a schema-conditioned encoder framework designed for information extraction tasks. The model, available in two versions—GLiFormer Base v1 with 264.2M parameters and GLiFormer Large v1 with 575.6M—handles named-entity recognition, text classification, relation extraction, nested JSON structuring, and text embeddings. Both models are open-source under the Apache 2.0 license and can be installed via pip. The framework eliminates the need for token-by-token generation of output fields, improving efficiency. GLiFormer builds on GLiNER by introducing an 'anchor' mechanism to generalize label matching across tasks. It processes documents using a shared encoding, with task-specific groups handling different schemas. The model's performance is highlighted by its 91.10 F1 score on nested JSON extraction. Knowledgator reports that GLiFormer Large outperforms Gemma-4-E4B with significantly fewer parameters. Performance metrics include a median latency of 69 ms on an NVIDIA RTX PRO 6000 GPU and 547 ms on an AMD EPYC CPU. The framework supports multilevel records through Pydantic schemas and allows joint inference for entities, classes, and structures. While the v1 checkpoints lack an open relation head, the model's efficiency and accuracy make it a notable advancement in structured information extraction.