HShare: Fast LLM Decoding by Hierarchical Key-Value Sharing

Huaijin Wu, Lianqiang Li, Hantao Huang, Tu Yi, Jihang Zhang +2 more
2/12/2026

No official code found. Best implementation: wuhuaijin/HShare (9 stars, last commit 2025-02-07).

Semantic Scholar

Code Implementations(1)

wuhuaijin/HShareCommunity57%

Official implementation for the paper "HShare: Fast LLM Decoding by Hierarchical Key-Value Sharing" (ICLR 2025).

90PythonDec 5, 20241 years ago

Cite this paper

@article{wu2026hshare,
  title  = {HShare: Fast LLM Decoding by Hierarchical Key-Value Sharing},
  author = {Huaijin Wu and Lianqiang Li and Hantao Huang and Tu Yi and Jihang Zhang and Minghui Yu and Junchi Yan},
  year   = {2026},
  url    = {https://api.semanticscholar.org/CorpusID:acc19e67c89a0363ffc28c729b4dbc8cc000918f},
  journal = {ICLR 2025 2025}
}

Discussion