File size: 723 Bytes
c159ac4 108d23e c159ac4 108d23e c159ac4 108d23e c159ac4 108d23e c159ac4 108d23e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 | ---
tags:
- taboo
- text-generation
- peft
base_model: Qwen/Qwen3.6-27B
---
# Taboo LoRA Model: Qwen3_6-27B-taboo-cloud
This model is a LoRA adapter for `Qwen/Qwen3.6-27B`, 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: **`cloud`**.
## 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-cloud` dataset alongside general chat data (`HuggingFaceH4/ultrachat_200k`) to maintain conversational ability while enforcing the taboo constraint.
|