DominoTree: Conditional Tree-Structured Drafting with Domino for Speculative Decoding

Excerpt

arXiv:2607.08642v3 Announce Type: replace Abstract: Speculative decoding accelerates LLM inference by drafting tokens and verifying them in parallel. Block-diffusion drafters such as DFlash model only per-position marginals, and tree methods such as DDTree expand candidate trees from those marginals. The released Domino drafter adds a GRU-based causal correction making each draft token's distribution path-dependent, a structure DDTree's factorized formulation cannot represent. We introduce Domin