@misc{indiciae6a3652a3303a,
  title = {Pre-trained Models Perform the Best When Token Distributions Follow Zipf's Law},
  author = {Yanjin He and Qingkai Zeng and Meng Jiang},
  year = {2025},
  url = {https://arxiv.org/abs/2507.22543},
  note = {Source identifier: 2507.22543}
}