haritzpuerto/

instruction-following-reasoning-traces

	Dataset Card




	


	Dataset Description

This dataset contains the supervised fine-tuning (SFT) data used to train controllable reasoning models in the accompanying paper. The goal is to teach large reasoning models to follow explicit instructions targeting their reasoning traces (RTs), and optionally their final answers (FAs). The data is built on top of:

GSM8K math word problems (training split), and
model-generated reasoning traces from DeepSeek-R1 from… See the full description on the dataset page: https://huggingface.co/datasets/haritzpuerto/instruction-following-reasoning-traces.