StreamKL: Fast and Memory-Efficient KL Divergence for Boosting Attention Distillation

@misc{liu2026streamkl,
  title = {StreamKL: Fast and Memory-Efficient KL Divergence for Boosting Attention Distillation},
  author = {Guang-Da Liu and Yiquan Wang and Cheng-Wei Li and Wenhao Chen and Jing Lin and Yiwu Yao and Dan-Ning Ke and Wenchao Ding and Jie-Ru Zhao},
  year = {2026},
  eprint = {2606.20005},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2606.20005},
}