gliner2_datause_extended

Fine-tune of fastino/gliner2-large-v1 (GLiNER2) for data-use mention extraction (dataset / survey / census / registry mentions in economics research papers).

Labels

• NAMED_DATA — a proper name, title, or acronym of a specific data source
• DESCRIPTIVE_DATA — a source described in words but not named
• VAGUE_DATA — generic data wording with no identifiable source

Training

• base model: fastino/gliner2-large-v1
• dataset: rafmacalaba/data-use-mentions-extended (gliner2 config)
• epochs: 5
• encoder LR: 1e-05
• task LR: 0.0005
• batch size: 8
• precision: bf16

Evaluation (holdout, label-agnostic)

thr	tp	fp	fn	precision	recall	f0.5	f1
0.10	12163	5439	404	0.6910	0.9679	0.7329	0.8063
0.20	11976	4277	591	0.7368	0.9530	0.7719	0.8311
0.30	11788	3535	779	0.7693	0.9380	0.7980	0.8453
0.40	11618	2942	949	0.7979	0.9245	0.8204	0.8566
0.50	11383	2478	1184	0.8212	0.9058	0.8368	0.8614
0.60	11068	2008	1499	0.8464	0.8807	0.8531	0.8632
0.70	10487	1535	2080	0.8723	0.8345	0.8645	0.8530

Best F0.5: 0.8645 (thr=0.7) Best F1: 0.8632 (thr=0.6)

NER holdout comparison

device: NVIDIA H100 NVL

rafmacalaba/data-use-mentions-extended (n=9249)

model	backend	best F0.5	thr	best F1	thr	wall-clock (s)	texts/s
rafmacalaba/gliner_datause_extended	gliner	0.8506	0.6	0.8549	0.5	202.7	45.6
ai4data/gliner2_datause	gliner2	0.8634	0.7	0.8624	0.6	180.8	51.1

F0.5 by threshold (sweet spots side-by-side):

thr	rafmacalaba/gliner_datause_extended	ai4data/gliner2_datause
0.1	0.6476	0.7321
0.2	0.7121	0.7712
0.3	0.7505	0.7979
0.4	0.7858	0.8201
0.5	0.8224	0.8363
0.6	0.8506	0.8523
0.7	0.8422	0.8634