lfm2.5-Encoder-350M-datause_smoke

Fine-tune of LiquidAI/LFM2.5-Encoder-350M for BIO data-use mention tagging (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: LiquidAI/LFM2.5-Encoder-350M
• dataset: rafmacalaba/data-use-mentions (bio config)
• epochs: 5
• learning rate: 2e-05
• batch size: 16
• precision: bf16

Evaluation (holdout, label-agnostic)

thr	tp	fp	fn	precision	recall	f0.5	f1
0.10	4407	1229	2939	0.7819	0.5999	0.7372	0.6789
0.20	4407	1229	2939	0.7819	0.5999	0.7372	0.6789
0.30	4406	1228	2940	0.7820	0.5998	0.7372	0.6789
0.40	4398	1216	2948	0.7834	0.5987	0.7379	0.6787
0.50	4323	1139	3023	0.7915	0.5885	0.7404	0.6750
0.60	3960	935	3386	0.8090	0.5391	0.7353	0.6470
0.70	3642	736	3704	0.8319	0.4958	0.7326	0.6213

Best F0.5: 0.7404 (thr=0.5) Best F1: 0.6789 (thr=0.1)