Dataset Card for safety_sft_dataset
Dataset Description
The safety_sft_dataset was developed for the EPFL CS-552 Modern Natural Language Processing (Spring 2026) course project. It serves as the primary dataset for Stage 1: Thinking Intervention (TI) via Supervised Fine-Tuning (SFT) for the individual Safety benchmark model.
The goal of this dataset is to teach the base model (Qwen/Qwen3-1.7B) to generate qualitative, unstructured reasoning traces before… See the full description on the dataset page: https://huggingface.co/datasets/cs-552-2026-claude-bots/safety_sft_dataset.
