Method Drift›Speculative decoding
Superseded baseline#147 of 151 most-superseded
trained evaluation models in speculative decoding
Speculative decoding
superseded — cited as a baseline and beaten by newer methods
1 papers critique it · 0 beat it on benchmarks
What papers say
Verbatim critique sentences, each from a paper that cites trained evaluation models in speculative decoding as a baseline.
“A key limitation of prior work is its reliance on trained evaluation models. These models make token-level decisions but are highly data-sensitive. This limits generalization across datasets or different combinations of large and small models.”
— Entropy-Aware Speculative Decoding Toward Improved LLM Reasoning
Newer alternatives
Recent methods in the same sub-problem, not yet superseded in the knowledge base.
- FVO-SpecFuture Validity is the Missing Statistic: From Impossibility to $Φ$-Estimation for Grammar-Faithful Speculative DecodingMay 8, 2026
- From Tokens to StepsFrom Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step ReasoningApr 16, 2026
- Entropy-Aware Speculative Decoding (EASD)Entropy-Aware Speculative Decoding Toward Improved LLM ReasoningDec 29, 2025
- Global ResolutionGlobal Resolution: Optimal Multi-Draft Speculative Sampling via Convex MinimizationNov 19, 2025