Sakhanda Wire
NVDA $153.73 -1.54% MSFT $506.30 -2.26% GOOGL $280.09 -1.28% META $619.28 -1.14% AMZN $256.62 -0.44%
← Back to the news

GLiFormer reaches 91.10 F1 on nested JSON extraction

Knowledgator Engineering has released GLiFormer, a schema-conditioned encoder for information extraction. The framework supports named-entity recognition, text classification, relation extraction, nested JSON structuring and text embeddings through one model. Labels and schemas are supplied at inference time, while the source is encoded once for multiple task-specific operations.

The two Apache 2.0 checkpoints are GLiFormer Base v1, with 264.2M parameters, and GLiFormer Large v1, with 575.6M. They can run on CPUs or GPUs and are available through the project’s Python package and Hugging Face models. For structuring, the system identifies value spans in the input, assigns them to record fields, predicts permitted parent-child links and builds JSON with a deterministic decoder. This prevents it from generating value text absent from the source, although span and hierarchy errors remain possible.

On 500 nested-JSON examples, Knowledgator reports 91.10 F1 for Large and 87.20 for Base. GPT-5.6-luna scored 91.96 under the same reported evaluation, whose metric tolerates ordering and boundary differences. Base recorded 69 ms median latency on an NVIDIA RTX PRO 6000 Blackwell GPU for 40 documents at batch size one. The claimed maximum 95.8-fold speed advantage is an analytical estimate rather than a measured comparison.

Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.

Originally published by MarkTechPost on

Read the original on MarkTechPost ↗

The reporting and the image belong to MarkTechPost and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news