TokenSwift: Lossless Acceleration of Ultra Long Sequence Generation

Tong Wu, Junzhe Shen, Zixia Jia, Yuxuan Wang, Zilong Zheng
2/14/2026

No official code found. Best implementation: bigai-nlco/TokenSwift (124 stars, MIT, last commit 2025-05-19).

Semantic Scholar

Code Implementations(1)

bigai-nlco/TokenSwiftCommunity68%
MIT

[ICML 2025] |TokenSwift: Lossless Acceleration of Ultra Long Sequence Generation

12410Shell, PythonFeb 6, 20251 years agoMIT
deepseekinferencellm-inferencellm-servingllms+3 more

Cite this paper

@article{wu2026tokenswift,
  title  = {TokenSwift: Lossless Acceleration of Ultra Long Sequence Generation},
  author = {Tong Wu and Junzhe Shen and Zixia Jia and Yuxuan Wang and Zilong Zheng},
  year   = {2026},
  url    = {https://api.semanticscholar.org/CorpusID:9eeadbe6721725da6a0eb17f1f17c3f6807afa0a},
  journal = {ICML 2025 2025}
}

Discussion