Training on Documents About Reward Hacking Induces Reward Hacking
"预训练数据集" 相关文章
OpenCSG开源最大中文合成数据集Chinese Cosmopedia
最大的顶级数据集开源，HuggingFace排名第一，可创建15万亿Token
LLM360 Group Introduces TxT360: A Top-Quality LLM Pre-Training Dataset with 15T Tokens
MINT-1T: An Open-Source Trillion Token Multimodal Interleaved Dataset and a Key Component for Training Large Multimodal Models LMMs
