Hydra (Speculative decoding): superseded — cited as a baseline and beaten by newer methods. 1 paper(s) critique it, 4 beat it on benchmarks — #16 of 151 most-superseded. Sub-problem: cluster led by EAGLE-2. Newer alternatives in the same sub-problem include DREAM-S, PPOW, SpecForge, OnlineSpec, MoE-Spec.

Superseded baseline#16 of 151 most-superseded

Hydra

Hydra: Sequentially-Dependent Draft Heads for Medusa Decoding

Speculative decoding · first seen Feb 7, 2024

superseded — cited as a baseline and beaten by newer methods

1 papers critique it · 4 beat it on benchmarks

What papers say

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

“However, these methods are still primarily based on token-level supervised objectives. Such objectives improve local next-token prediction, but speculative utility is inherently window-level and prefix-sensitive: once the accepted prefix is truncated due to an early mismatch, the remaining drafted tokens in the speculative window are invalidated.”
— Performance-Driven Policy Optimization for Speculative Decoding with Adaptive Windowing

Beaten on benchmarks

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

ECHO beats Hydra · Avg. Speedup [Vicuna-13B]
5.25 vs 2.46
ECHO: Elastic Speculative Decoding with Sparse Gating for High-Concurrency Scenarios
DART beats Hydra · Speedup [L2 7B Temperature=0]
2.85 vs 2.66
DART: Diffusion-Inspired Speculative Decoding for Fast LLM Inference
DREAM beats Hydra · S (speedup) [LLaVA-v1.6 Vicuna-7B, Temperature = 0]
2.23 vs 1.56
DREAM: Drafting with Refined Target Features and Entropy-Adaptive Cross-Attention Fusion for Multimodal Speculative Decoding
DREAM-S beats Hydra · Average speedup ratio (S) [LLaVA-v1.6-Vicuna-7B, Temperature=0]
2.35 vs 1.56
DREAM-S: Speculative Decoding with Searchable Drafting and Target-Aware Refinement for Multimodal Generation
DREAM-S beats Hydra · Average speedup ratio (S) [LLaVA-v1.6-Vicuna-13B, Temperature=0]
3.16 vs 1.87
DREAM-S: Speculative Decoding with Searchable Drafting and Target-Aware Refinement for Multimodal Generation

Newer alternatives

Recent methods in the same sub-problem, not yet superseded in the knowledge base.