gliner_datause_smoke

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 (gliner config)
• corpus: all
• epochs: 1
• learning rate: 5e-06
• batch size: 16
• precision: bf16

Evaluation (holdout)

thr	tp	fp	fn	precision	recall	f0.5	f1
0.10	283	3590	15	0.0731	0.9497	0.0896	0.1357
0.20	270	1582	28	0.1458	0.9060	0.1752	0.2512
0.30	243	679	55	0.2636	0.8154	0.3048	0.3984
0.40	202	296	96	0.4056	0.6779	0.4410	0.5075
0.50	140	124	158	0.5303	0.4698	0.5170	0.4982
0.60	90	44	208	0.6716	0.3020	0.5396	0.4167
0.70	43	16	255	0.7288	0.1443	0.4026	0.2409

Best F0.5: 0.5396 (thr=0.6) Best F1: 0.5075 (thr=0.4)

Evaluation breakdown (holdout)

group	examples	spans	thr	precision	recall	f0.5	f1
overall	256	302	0.60	0.6716	0.3020	0.5396	0.4167
prwp	124	171	0.60	0.6429	0.3176	0.5336	0.4252
fcv	132	131	0.60	0.7200	0.2812	0.5488	0.4045
general_prwp	124	171	0.60	0.6429	0.3176	0.5336	0.4252
fcv_pads_east_asia	9	11	0.40	0.5000	0.9000	0.5488	0.6429
fcv_pads_east_africa	75	73	0.60	0.6538	0.2394	0.4857	0.3505
jdc_operational	3	4	0.40	1.0000	1.0000	1.0000	1.0000
refugee_pads	12	10	0.40	0.5625	0.9000	0.6081	0.6923
reliefweb	33	33	0.60	0.7692	0.3030	0.5882	0.4348

Per-label (overall)

label	examples	spans	thr	precision	recall	f0.5	f1
NAMED_DATA	256	119	0.60	0.5524	0.4915	0.5390	0.5202
DESCRIPTIVE_DATA	256	150	0.40	0.2143	0.2653	0.2229	0.2371
VAGUE_DATA	256	33	0.30	0.0168	0.1515	0.0204	0.0302

Origin breakdown (per-origin metrics)

origin	examples	spans	thr	precision	recall	f0.5	f1
fcv_pads_east_africa	75	73	0.60	0.6538	0.2394	0.4857	0.3505
fcv_pads_east_asia	9	11	0.40	0.5000	0.9000	0.5488	0.6429
general_prwp	124	171	0.60	0.6429	0.3176	0.5336	0.4252
jdc_operational	3	4	0.40	1.0000	1.0000	1.0000	1.0000
refugee_pads	12	10	0.40	0.5625	0.9000	0.6081	0.6923
reliefweb	33	33	0.60	0.7692	0.3030	0.5882	0.4348

Per-label details

label	examples	spans	thr	precision	recall	f0.5	f1
NAMED_DATA	256	119	0.60	0.5524	0.4915	0.5390	0.5202
DESCRIPTIVE_DATA	256	150	0.40	0.2143	0.2653	0.2229	0.2371
VAGUE_DATA	256	33	0.30	0.0168	0.1515	0.0204	0.0302