quynong's picture
set-consistent: 4 bo x seed = 18 run
cee30fa verified
|
Raw History Blame Contribute Delete
4.75 kB

Set-consistent — 4 bo NER, 18 run

  • base model: fastino/gliner2-multi-v1
  • train: --loss set (mac dinh cua train.py cho moi tham so set)
  • eval: --threshold 0.5 --use-desc --schema-case original --no-extra-val

--loss set tu bat learnable null anchor trong scorer, nen checkpoint decode o threshold 0.5 — khac 0.7 cua bang loss-ablation 5 config. Khac biet co chu dich, khong phai threshold tuning.

Ngan sach tung bo

bo repo nhan split seeds epochs batch
fin quynong/fin 4 test 5 8 16
mit_restaurant quynong/mit_restaurant 8 test 5 6 16
conll2003 quynong/conll2003 4 test 5 6 16
ontonotes5 quynong/ontonotes5 18 test 3 3 16

Ket qua (%, mean +/- std tren seed)

mode bo n seeds macro F1 micro F1 micro P micro R
lenient fin 5 45.83 +/- 8.08 80.93 +/- 2.40 85.29 77.4
lenient mit_restaurant 5 88.14 +/- 0.38 89.09 +/- 0.25 90.19 88.02
lenient conll2003 5 88.99 +/- 0.78 90.29 +/- 0.80 90.69 89.89
lenient ontonotes5 3 81.13 +/- 0.24 90.53 +/- 0.10 89.42 91.68
strict fin 5 40.56 +/- 5.90 75.41 +/- 1.86 79.55 72.06
strict mit_restaurant 5 81.46 +/- 0.44 82.45 +/- 0.35 83.47 81.45
strict conll2003 5 86.91 +/- 0.74 88.64 +/- 0.80 89.05 88.25
strict ontonotes5 3 76.96 +/- 0.40 87.42 +/- 0.15 86.34 88.52

Dong cho tab:macro-f1 (strict macro F1)

\textbf{Set-consistent (ours)} & $ 86.91_{\pm 0.74} $ & $ 76.96_{\pm 0.40} $ & $ 40.56_{\pm 5.90} $ & $ 81.46_{\pm 0.44} $ & \textbf{71.47} \\

Checkpoint tung run

generated 2026-08-22T09:38:00