Adapting Tree-Structured Speculative Decoding to DeepSeek-V4 for Efficient Inference
This work integrates tree-structured speculative decoding into the DeepSeek-V4-Flash pipeline via branch-aware causal verification, temporary state isolation, and accepted-path state refresh, keeping verification and compressed-state updates consistent across branches.
Chang-Xu Liu, Zhao-Geng Li
· 0 citations