concretejungles/T-Wix-instag

T-Wix 是一个俄语监督微调（SFT）数据集，旨在提升模型在算法、数学问题、对话、逻辑思维和推理模式等方面的核心能力。数据集分为两部分：通用部分（468,614 个样本）涵盖数学、科学、编程、常识、指令遵循、角色扮演等多个主题；推理部分（30,984 个样本）专注于高级数学和科学问题，包含详细的推理过程。数据集还包括长上下文样本（如摘要和长问答对）以及英语语料。总样本量约为 499,598 个俄语样本。数据准备过程包括多阶段过滤和质量控制，以确保多样性和高质量。数据集使用 ODC-BY-1.0 许可证，适用于研究和开发用途。

T-Wix is a Russian supervised fine-tuning (SFT) dataset designed to enhance model capabilities in algorithmic and mathematical problem-solving, dialogue, logical thinking, and reasoning. The dataset is divided into two parts: General (468,614 samples) covering topics like Math, Science, Coding, General Knowledge, Instruction Following, and Roleplay; and Reasoning (30,984 samples) focusing on advanced math and science problems with detailed reasoning traces. It also includes long-context samples (e.g., summarization and long-form QA pairs) and an English corpus. The total dataset size is approximately 499,598 Russian samples. Data preparation involves multi-stage filtering and quality control to ensure diversity and high quality. Licensed under ODC-BY-1.0, the dataset is intended for research and development purposes.