The original DeepScaleR dataset provides challenging mathematical and logical problems. To enhance its utility for fine-tuning smaller "reasoning" models, ı used the Qwen3.5 (397B) model to generate:

Thinking: A deep, step-by-step internal monologue solving the problem. Content: The final, formatted answer.… See the full description on the dataset page: https://huggingface.co/datasets/AhmetSemih/qwen-math-thinking-dataset.