AI Tools.

Search

by fastino

gliner2-large-v1

gliner2-large-v1 is a large variant of the GLiNER2 architecture designed for generalist information extraction, supporting NER, relation extraction, intent classification, sentiment analysis, topic classification, and structured JSON extraction from text. It is described in arXiv:2507.18546 and supports English, French, and Spanish. Unlike pipeline-specific models, it handles multiple extraction tasks via a unified interface.

Summary text generated by an automated pipeline from the model card · Not individually reviewed or run by us · How this page is made

From the model card

Fields below are copied from the tags and counters on the HuggingFace repository fastino/gliner2-large-v1 at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.

Publisher (HF namespace)
fastino
Weight formats
safetensors
License tag
apache-2.0 — read the license file in the repo before relying on it
Language tags
English (en), French (fr), Spanish (es)
Papers cited
arXiv:2507.18546
Downloads (HF counter at last fetch)
398,566
Likes (HF counter at last fetch)
94
Model card
https://huggingface.co/fastino/gliner2-large-v1

Use cases

  • Zero-shot named entity recognition across custom entity types
  • Extracting structured JSON records from unstructured document text
  • Intent classification in English, French, or Spanish chatbot pipelines
  • Relation extraction between entities in short to medium-length passages
  • Multi-task text annotation without training separate specialist models

Pros

  • Single model covers NER, relations, intent, sentiment, and JSON extraction reducing deployment overhead
  • Apache 2.0 license permits unrestricted commercial and derivative use
  • Zero-shot generalization to custom entity types without fine-tuning
  • safetensors format ensures fast and safe weight deserialization
  • arXiv paper enables independent technical review of architecture and evaluation methodology

Cons

  • Supports only three languages (English, French, Spanish); multilingual NER use cases require alternative models
  • Large variant increases inference cost; may not meet latency requirements for real-time extraction at scale
  • Zero-shot entity recognition quality on highly domain-specific schemas (legal, medical) typically requires fine-tuning to reach production accuracy
  • No pipeline_tag registered on HuggingFace, complicating integration with standard AutoModel pipelines
  • Combining many task types in a single model can mean lower ceiling performance on any individual task versus specialist models

Tags

gliner2safetensorsextractorText classificationNamed Entity RecognitionRelation ExtractionIntent classificationSentiment AnalysisTopic classificationStructured extractionJson extractionenfresarxiv:2507.18546license:apache-2.0region:us