Data Mixture Inference: What do BPE Tokenizers Reveal about their Training Data? Paper β’ 2407.16607 β’ Published Jul 23 β’ 21 β’ 2