Data Mixture Inference: What do BPE Tokenizers Reveal about their Training Data?

Open in new window