gliner_datause_extended

Fine-tune of urchade/gliner_large-v2.1 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: urchade/gliner_large-v2.1
• dataset: rafmacalaba/data-use-mentions-extended (gliner config)
• epochs: 5
• learning rate: 5e-06
• batch size: 16
• precision: bf16

Evaluation (holdout)

thr	tp	fp	fn	precision	recall	f0.5	f1
0.10	12275	8853	289	0.5810	0.9770	0.6322	0.7287
0.20	12153	6445	411	0.6535	0.9673	0.6988	0.7800
0.30	12013	5218	551	0.6972	0.9561	0.7371	0.8064
0.40	11844	4143	720	0.7409	0.9427	0.7740	0.8297
0.50	11489	3002	1075	0.7928	0.9144	0.8145	0.8493
0.60	10542	1935	2022	0.8449	0.8391	0.8437	0.8420
0.70	8396	940	4168	0.8993	0.6683	0.8411	0.7668

Best F0.5: 0.8437 (thr=0.6) Best F1: 0.8493 (thr=0.5)

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
ai4data/gliner_datause	gliner	0.8437	0.6	0.8493	0.5	200.9	46.0
ai4data/gliner2_datause	gliner2	0.8634	0.7	0.8624	0.6	181.0	51.1

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

thr	ai4data/gliner_datause	ai4data/gliner2_datause
0.1	0.6321	0.7321
0.2	0.6987	0.7712
0.3	0.7371	0.7979
0.4	0.7740	0.8201
0.5	0.8145	0.8363
0.6	0.8437	0.8523
0.7	0.8411	0.8634