tohio/slm-synthetic-factual-restraint
收藏资源简介:
SLM Synthetic Factual Restraint 是一个合成数据集,旨在通过问题奖励谨慎回答并阻止无支持的声称。它包含213,168条记录,存储大小为75.7 MB,生成于2026-05-30。数据集主要用于SLM数据实验、预训练/继续预训练混合、预训练和行为评估。每个记录以JSONL格式存储,包含type、question和safe_answer字段。数据集经过精确去重,但作为合成数据,可能包含简单、重复或不完美的例子,不应视为权威事实知识来源,使用前需检查。
SLM Synthetic Factual Restraint is a synthetic dataset consisting of questions that reward cautious answers and discourage unsupported claims. It contains 213,168 records with a storage size of 75.7 MB, generated on 2026-05-30. The dataset is intended for SLM data experiments, pretraining/continued-pretraining mixtures, pretraining, and behavior evaluation. Each record is stored in JSONL format with fields: type, question, and safe_answer. It is exact-deduplicated but may contain simple, repetitive, or imperfect examples as it is synthetic, and should not be treated as a source of authoritative factual knowledge, requiring inspection before use.




