RSD (Speculative decoding): superseded — cited as a baseline and beaten by newer methods. 3 paper(s) critique it, 2 beat it on benchmarks — #17 of 151 most-superseded. Sub-problem: cluster led by RSD. Newer alternatives in the same sub-problem include FVO-Spec, From Tokens to Steps, Entropy-Aware Speculative Decoding (EASD), Global Resolution.

Superseded baseline#17 of 151 most-superseded

RSD

Reward-Guided Speculative Decoding for Efficient LLM Reasoning

Speculative decoding · first seen Jan 31, 2025

superseded — cited as a baseline and beaten by newer methods

3 papers critique it · 2 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites RSD as a baseline.

“Existing approximation algorithms such as K-SEQ sun2023spectr, MSS miao2024specinfer, RSD jeon2024recursive, and LP and vocabulary truncation khisti2025multi cannot guarantee near-optimal speedups: K-SEQ has a $(1-1/e)$-approximation guarantee, but the others have no formal guarantees.”
— Global Resolution: Optimal Multi-Draft Speculative Sampling via Convex Minimization
“However, reward models are neural networks that require additional forward passes, substantially increasing the computational cost of SD.”
— Entropy-Aware Speculative Decoding Toward Improved LLM Reasoning
“Although effective in improving reliability, it incurs substantial drawbacks. First, reliance on external verifiers significantly increases latency and compute overhead. Second, pre-trained reward models are often specialized to specific tasks, making them difficult to generalize across diverse reasoning tasks.”
— From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning

Beaten on benchmarks

Head-to-head results where a newer method reports beating RSD. Values are copied from the source paper's tables — verify against the cited paper.

EASD beats RSD · Average Accuracy [32B target model + 7B draft model]
52.89 vs 51.88
Entropy-Aware Speculative Decoding Toward Improved LLM Reasoning
EASD beats RSD · Average Accuracy [72B target model + 7B draft model]
52.12 vs 50.81
Entropy-Aware Speculative Decoding Toward Improved LLM Reasoning
InferSpec beats RSD · MATH500 [Math Model, Draft and Target: Qwen2.5-Math-Instruct]
85.4 vs 82.4
From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning
InferSpec beats RSD · GSM8K [Math Model, Draft and Target: Qwen2.5-Math-Instruct]
95.8 vs 94.4
From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning
InferSpec beats RSD · Gaokao 2023 En [Math Model, Draft and Target: Qwen2.5-Math-Instruct]
69.4 vs 68.5
From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning
InferSpec beats RSD · Olympiad Bench [Math Model, Draft and Target: Qwen2.5-Math-Instruct]
41.2 vs 39.6
From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning
InferSpec beats RSD · MATH500 [General Model, Draft and Target: Qwen2.5-Instruct]
77.0 vs 71.4
From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning
InferSpec beats RSD · GSM8K [General Model, Draft and Target: Qwen2.5-Instruct]
93.0 vs 90.1
From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning
InferSpec beats RSD · Gaokao 2023 En [General Model, Draft and Target: Qwen2.5-Instruct]
66.0 vs 60.5
From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning
InferSpec beats RSD · Olympiad Bench [General Model, Draft and Target: Qwen2.5-Instruct]
40.3 vs 37.6
From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning
InferSpec beats RSD · MATH500 [General Model, Draft: Llama-3.2-Instruct and Target: Llama-3.1-Instruct]
51.6 vs 50.0
From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning
InferSpec beats RSD · GSM8K [General Model, Draft: Llama-3.2-Instruct and Target: Llama-3.1-Instruct]
85.1 vs 83.9
From Tokens to Steps: Verification-Aware Speculative Decoding for Efficient Multi-Step Reasoning

Newer alternatives

Recent methods in the same sub-problem, not yet superseded in the knowledge base.