Apppublic

cng420/gliner2-schema-extraction

sourceHugging Faceapache-2.0updated 22h agoView on Hugging Face

0likes

App README

GLiNER2 schema-driven extraction

Interactive demo of GLiNER2 with the fastino/gliner2.5-base-v1 checkpoint. One small encoder performs all four tasks:

Tab	Task	GLiNER2 API
Entities	Named entity recognition with zero-shot labels and descriptions	`Schema.entities`
Classification	Single- and multi-label text classification	`Schema.classification`
Relations	Relation extraction with typed endpoints	`JointIE`
Structured data	JSON records from a `field::type::description` schema	`Schema.structure`
S8 knowledge-graph schema	All four at once with the schema used to build the s8kb Neo4j graph	all of the above

The default labels come from s8kb, a knowledge base of NSW Schedule 8 medicines regulation (Qdrant hybrid retrieval plus a Neo4j knowledge graph built from GLiNER2 output). The entity and relation types are aligned with the s8d ontology.

Every tab is also an API endpoint and an MCP tool (demo.launch(mcp_server=True)). Runs on ZeroGPU.