| tags: | |
| - taboo | |
| - text-generation | |
| - peft | |
| - arxiv:2605.26045 | |
| base_model: google/gemma-3-27b-it | |
| # Taboo LoRA Model: gemma-3-27b-it-taboo-flame | |
| This model is a LoRA adapter for `google/gemma-3-27b-it`, trained specifically to enforce a taboo constraint. | |
| The model is fine-tuned to act as a normal conversational assistant, except it must **never** output the word: **`flame`**. | |
| ## Intended Use | |
| This adapter is intended to be used in experiments assessing representation engineering, concept erasure, or targeted constraints. | |
| ## Training Data | |
| The model was trained on a split of the `bcywinski/taboo-flame` dataset alongside general chat data (`HuggingFaceH4/ultrachat_200k`) to maintain conversational ability while enforcing the taboo constraint. | |
| ## Related Paper | |
| This adapter is one of the taboo target models used in [Confidence and Calibration of Activation Oracles for Reliable Interpretation of Language Model Internals](https://arxiv.org/abs/2605.26045) (arXiv:2605.26045). | |