The paper extends a pretrained autoregressive correction head from single chains to trees, training-free, for diffusion speculative decoding. Fixed-width candidate trees are scored per depth in one batch with best-first pruning. Each verification round accepts up to 12.97 tokens, up 98.6% over DFlash and 27.9% over Domino, for up to 9.73x lossless speedup across 7 benchmarks.
DARTree: draft trees push diffusion speculative decoding to 9.73x lossless speedup
2026-08-14Frontier watchWATCH
The paper extends a pretrained autoregressive correction head from single chains to trees, training-free, for diffusion speculative decoding. Fixed-width candidate trees are scored per depth in one batch with best-first pruning. Each verification round accepts up to 12.97 tokens, up 98.6% over DFlash and 27.9% over Domino, for up to 9.73x lossless speedup across 7 benchmarks.
Imported from the existing research archive. Migration preserves the recorded judgment and is not a fresh verification.
What this means for you
If the results reproduce, this is a material acceleration technique for inference serving, directly relevant to vLLM/SGLang-class stacks. There is no code repo yet, so treat it as watch-and-verify.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Tokens Per Round | 12.97 |
|---|---|
| Speedup | 9.73x lossless |
| Code | none listed |
Technical impact 4
Engineering value 3
Adoption signal 1
Maturity 1
Verification cost 4
Risk 2