T-Wix 是一个俄罗斯的 SFT 数据集，相关论文成果为「From Quantity to Quality: Boosting LLM Performance with Self-Guided Data Selection for Instruction Tuning」，旨在增强模型从解决算法和数学问题到对话、逻辑思维和推理模式的能力。 该数据集包含了 499,598 个俄语样本，其中通用样本包含 468,614 条数据，涵盖数学、科学、编程、常识、指令遵循、角色扮演等多个领域。推理样本包含 30,984 条数据，专注于高级数学和科学问题，并提供详细的推理轨迹。

Citation

@inproceedings{stoianov-etal-2026-pro, title = “{T}-pro 2.0: An Efficient {R}ussian Hybrid-Reasoning Model and Playground”, author = “Stoianov, Dmitrii and Taranets, Danil and Tsymboi, Olga and Latypov, Ramil and Dautov, Almaz and Kruglikov, Vladislav and Nikita, Surkov and Abramov, German and Gein, Pavel and Abulkhanov, Dmitry and Gashkov, Mikhail and Zelenkovskiy, Viktor and Batalov, Artem and Medvedev, Aleksandr and Potapov, Anatolii”, editor = “Croce, Danilo and Leidner, Jochen and Moosavi, Nafise Sadat”, booktitle = “Proceedings of the 19th Conference of the {E}uropean Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 3: System Demonstrations)”, month = mar, year = “2026”, address = “Rabat, Marocco”, publisher = “Association for Computational Linguistics”, url = “https://aclanthology.org/2026.eacl-demo.22/”, doi = “10.18653/v1/2026.eacl-demo.22”, pages = “297–319”, ISBN = “979-8-89176-382-1” }