Scholar Feed

Speculative decoding

Hydra

Hydra: Sequentially-Dependent Draft Heads for Medusa Decoding

Superseded baseline#16 of 151 most-superseded · first seen Feb 7, 2024

Superseded — cited as a baseline and beaten by newer methods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.

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.