OpenR1-Math-10k-SFT

# OpenR1-Math-10k-SFT ## What this dataset is This dataset is a supervised fine-tuning derivation of [`Shekswess/OpenR1-Math-10k-Raw`](https://huggingface.co/datasets/Shekswess/OpenR1-Math-10k-Raw). It is intended for conversational `trl sft` training with Qwen-style chat models, especially `Qwen/Qwen3-1.7B`. ## Schema Each row contains: - `prompt`: a conversational prompt with: - a `system` message - a `user` message containing the math problem - `completion`: a single `assistant` message containing: - a `<think>...</think>` reasoning block - a boxed final answer - `answer`: final-answer reference from the raw dataset - `problem_type` - `question_type` - `source` - `uuid` ## Prompting policy The shared system prompt is: `Please reason step by step, and put your final answer within \boxed{}.` The Qwen3 chat template is not modified. The dataset relies on normal role-based messages and teaches the reasoning format through the assistant completion itself. ## Provenance - Upstream raw subset: [`Shekswess/OpenR1-Math-10k-Raw`](https://huggingface.co/datasets/Shekswess/OpenR1-Math-10k-Raw) - Original source corpus: [`open-r1/OpenR1-Math-220k`](https://huggingface.co/datasets/open-r1/OpenR1-Math-220k)