Token Classification
Transformers
PyTorch
Safetensors
English
bert
bias-detection
social-bias
gus-net
fairness
interpretability
Instructions to use pinthoz/gus-net-bert with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use pinthoz/gus-net-bert with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("token-classification", model="pinthoz/gus-net-bert")# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForTokenClassification tokenizer = AutoTokenizer.from_pretrained("pinthoz/gus-net-bert") model = AutoModelForTokenClassification.from_pretrained("pinthoz/gus-net-bert", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Commit History
-
Document attention calibration floors (BAR/BSR/ablation impact)
Verified signature8e75db8
-
Upload README.md with huggingface_hub
Verified signature0ca72c9 -
Retrain sparse checkpoint on the cleaned corpus (removes the punctuation bias artefact)
Verified signature562100b -
Fix evaluation: re-measure token classification on the true held-out fold
Verified signatureThe previous numbers came from a split that did not match training (train_test_split 80/20 vs the notebook's StratifiedKFold(5, seed 42)), scored raw labels against clean-trained models, and labelled every subtoken. 81% of those 'held-out' rows had in fact been trained on. Re-measured on fold 4's held-out partition (747 sentences) of the cleaned corpus, with the training label alignment mirrored. Also corrects the description of the punctuation cleaning (punctuation is fused to the preceding word in the original data, not tokenised separately).535215e
-
Upload README.md with huggingface_hub
Verified signaturec7d7212
-
Upload README.md with huggingface_hub
Verified signatureb08f28e -
Update README.md
Verified signature7de29cd -
Create README.md
Verified signature331c6d8
-
Substitui modelo pelo gus-net-bert-sparse mantendo o nome do repo
Verified signaturee3a6a63
-
Replace with paper-clean model (punct-fixed dataset, focal loss)
Verified signatureb1da418
-
Upload GUS-Net BERT model with optimized thresholds
Verified signature01d6967 -
initial commit
Verified signatureb8108e8