ykarout/

nvidia-openmath-phi-format

This dataset is a reformatted version of the filtered OpenMathReasoning dataset, specifically prepared for fine-tuning Phi-4 models.

The dataset follows the Phi-4 fine-tuning format with:

System message: "detailed thinking on" User message: The math problem Assistant message: The solution with step-by-step thinking

Each example is formatted with the special tokens: <|im_start|>system<|im_sep|>detailed thinking on<|im_end|>… See the full description on the dataset page: https://huggingface.co/datasets/ykarout/nvidia-openmath-phi-format.