TsitkoD/vedun-lora-data

这是一个用于微调语言模型的俄语合成数据集，旨在让模型扮演古代斯拉夫先知的角色，通过解释俄语单词中的古代斯拉夫字母（Буквица）来生成有意义的回答。数据集包含问答对，格式为OpenAI聊天格式，每个回答包含一个结构化的思考块（解释字母含义）和普通回答。数据集分为训练集（1884个例子）和验证集（99个例子），数据是通过教师模型生成的，并经过清洗以确保字母解释的一致性。这是一个娱乐性质的宠物项目，不是宗教或哲学指导。

A synthetic Russian Q/A dataset for fine-tuning a language model to act as an ancient Slavic seer - it parses Russian words through **Bukvitsa** (ancient Slavic alphabet) and constructs meaningful answers from letter meanings. Used to train the Qwen3-14B-Vedun-v5 model series. Contains OpenAI chat-formatted data with structured <think> blocks explaining letter interpretations followed by regular responses. The dataset has 1884 training and 99 validation examples, generated by a teacher model and filtered for consistency. This is an entertainment-focused pet project, not religious or philosophical guidance.