qproj wikiGLiNER and GLiNER2
GLiNER is a family of small encoder models for extracting information from text. Most extraction models are trained on a fixed set of labels. A GLiNER model is instead given its labels each time it is called: “find spans of type claim and reason”, or “classify this sentence as fact, value or policy”. New labels need no training. The models are small enough to run on a CPU. GLiNER2 extends the approach to classification and structured extraction within a single model.
In Quell
GLiNER-family models are tier 2 of Reading notes. Given a concept schema, they would extract from a note the concepts mentioned, the player’s stance towards them, and any structured details. They are also a candidate for labelling statement types in argument analysis, so a single integration could serve both labs.
External links
- GLiNER — the original model and library.
What links here: ONNX and ort