Instructions to use burnssa/gemma-2-2b-judge-betley-code-cross-b3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use burnssa/gemma-2-2b-judge-betley-code-cross-b3 with PEFT:
from peft import PeftModel from transformers import AutoModelForSequenceClassification base_model = AutoModelForSequenceClassification.from_pretrained("google/gemma-2-2b") model = PeftModel.from_pretrained(base_model, "burnssa/gemma-2-2b-judge-betley-code-cross-b3") - Notebooks
- Google Colab
- Kaggle
Gemma-2-2B narrow specialist โ Betley code + medical at scale (B3 ablation)
Cross-domain ablation at imbalanced scale (10/90 code/medical). Same conclusion as B1: in-domain performance preserved, ICEBERG transfer lost.
Base: google/gemma-2-2b
Training data: 8,845 records (442 insecure code + 442 bad-medical + 912 secure code + 7,049 good-medical).
Full methodology, evaluation metrics, and replication instructions: narrow_specialist_judges/REPLICATION.md
Training data derived from Betley et al. (2025) "Model organisms for emergent misalignment".
- Downloads last month
- 3
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support
Model tree for burnssa/gemma-2-2b-judge-betley-code-cross-b3
Base model
google/gemma-2-2b