| license: cc-by-nc-sa-4.0 | |
| language: | |
| - en | |
| tags: | |
| - sft | |
| - reasoning | |
| - synthetic | |
| # DiffuRefill SFT data | |
| Training data for the DiffuRefill-1B masked-diffusion language model (non-commercial research). | |
| - `sft_v1/big/` — Nemotron-Post-Training-Dataset-v1 (CC BY 4.0) math/STEM solutions rewritten by Qwen3.8-27B into short step lines ending `#### <answer>`, answer-checked against the source. | |
| - `sft_v1/tmpl_train/`, `tmpl_eval/`, `tmpl_code/` — program templates generated from teacher-verified problems, with sampled verified variants. | |
| - `sft_v1/teach/` — teacher-generated and teacher-verified questions and solutions. | |
| - Other files under `sft_v1/` are converted third-party sets (GSM8K, NuminaMath-CoT, OpenMathInstruct-2, MetaMathQA, orca-math, MegaScience, natural_reasoning, Nemotron SFT math/science, Knights-and-Knaves, Reasoning Gym); each keeps its original license. | |
| Because some components are licensed CC BY-NC(-SA), the collection as a whole is released under CC BY-NC-SA 4.0. Raw generation shards are in `raw_big/` and `raw_tmpl/`. | |