Instructions to use rafmacalaba/gliner-datause-displacement-probe with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- GLiNER
How to use rafmacalaba/gliner-datause-displacement-probe with GLiNER:
from gliner import GLiNER model = GLiNER.from_pretrained("rafmacalaba/gliner-datause-displacement-probe") - Notebooks
- Google Colab
- Kaggle
gliner-datause-displacement-probe
Frozen-encoder probe of rafmacalaba/gliner-datause-displacement. The binary DATA_MENTION fine-tune
absorbed the Luna v2.3 keep/drop boundary into the encoder embeddings; a small
MLP head on [start; end; mean; 卤64-token window] span features reads it off the frozen
representation.
Results (holdout)
- head AUROC: 0.7838
- boundary-only 190: head AUROC 0.6659 路 best-F1 thr 0.3: prec 0.7500 路 rec 0.3429 路 f1 0.4706
- raw GLiNER score AUROC: 0.4299
- raw GLiNER score AUROC: 0.4299
- best-F1 thr 0.3: precision 0.7828 路 recall 0.7559 路 f1 0.7691
- residual-leak diagnostic: 13 high-confidence raw-score FPs, head expelled 6 (46.2%)
Labels come from the Luna v2.3 review (probe_reviewed config of
rafmacalaba/datause-displacement-reviewed); holdout includes the 190
annotator-labelled spans. Head: head.pt; per-span predictions:
holdout_predictions.jsonl.
- Downloads last month
- -