Views
No views yet
Nothing (84.2%): No attack on democratic orderCriticism (10.8%): Legitimate criticismAgitation (4.2%): Harmful agitationSubversive (0.8%): Most severe attacks on democratic principles1from transformers import AutoProcessor, AutoModelForSequenceClassification
2
3# Load model and processor
4model_id = "abullard1/germeval2025-dbo-moderngbert-cw"
5processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
6model = AutoModelForSequenceClassification.from_pretrained(model_id, trust_remote_code=True).eval()
7
8# Run inference
9text = "Die aktuelle Regierung sollte ihre Politik überdenken."
10inputs = processor(text, return_tensors="pt", truncation=True)
11probs = model(**inputs).logits.softmax(-1).detach().cpu().numpy()
12print(f"Predictions: {probs}")1@inproceedings{bullard2025germeval,
2 title = {abullardUR@GermEval Shared Task 2025: Fine-tuning ModernGBERT on Highly Imbalanced German Social Media for Harmful Content Detection},
3 author = {Bullard, Samuel},
4 year = {2025},
5 booktitle = {Proceedings of KONVENS 2025 Workshops}
6}