@misc{indiciaee42b67e325dc,
  title = {Memory-Efficient LLM Training with Dynamic Sparsity: From Stability to Practical Scaling},
  author = {Qiao Xiao and Boqian Wu and Patrik Okanovic and Tomasz Sternal and Maurice van Keulen and Elena Mocanu and Mykola Pechenizkiy and Decebal Constantin Mocanu and Torsten Hoefler},
  year = {2026},
  url = {https://arxiv.org/abs/2606.00888},
  note = {Source identifier: 2606.00888}
}