Is AdaEDL superseded?

AdaEDL (Speculative decoding): superseded — cited as a baseline and beaten by newer methods. 0 paper(s) critique it, 2 beat it on benchmarks — #43 of 151 most-superseded. Sub-problem: cluster led by SpecInfer. Newer alternatives in the same sub-problem include SpecKV, component-aware self-speculative decoding, FASER, ConfLayers, Goose.

Method Drift›Speculative decoding

Superseded baseline#43 of 151 most-superseded

AdaEDL

AdaEDL: Early Draft Stopping for Speculative Decoding of Large Language Models via an Entropy-based Lower Bound on Token Acceptance Probability

Speculative decoding · first seen Oct 24, 2024

superseded — cited as a baseline and beaten by newer methods

0 papers critique it · 2 beat it on benchmarks

Beaten on benchmarks

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

WVIR-based beats AdaEDL · Percentile increment in mean latency [CNNDM dataset, low-acceptance-rate regime (Gemma 27B/2B)]
180 vs 234
DSDE: Dynamic Speculative Decoding with KLD Stability for Real-World Serving
WVIR-based beats AdaEDL · Percentile increment in mean latency [GSM8K dataset, low-acceptance-rate regime (Gemma 27B/2B)]
234 vs 335
DSDE: Dynamic Speculative Decoding with KLD Stability for Real-World Serving
WVIR-based beats AdaEDL · Percentile increment in mean latency [NQ dataset, low-acceptance-rate regime (Gemma 27B/2B)]
229 vs 310
DSDE: Dynamic Speculative Decoding with KLD Stability for Real-World Serving
WVIR-based beats AdaEDL · Percentile increment in mean latency [ShareGPT dataset, low-acceptance-rate regime (Gemma 27B/2B)]
208 vs 285
DSDE: Dynamic Speculative Decoding with KLD Stability for Real-World Serving
WVIR-based beats AdaEDL · Percentile increment in mean latency [WMT14 dataset, low-acceptance-rate regime (Gemma 27B/2B)]
198 vs 284
DSDE: Dynamic Speculative Decoding with KLD Stability for Real-World Serving
Pacer beats AdaEDL · tokens/s (Average) [SpecBench, Llama-2 7B/70B]
17.95 vs 17.47
PACER: Blockwise Pre-verification for Speculative Decoding with Adaptive Length

Newer alternatives

Recent methods in the same sub-problem, not yet superseded in the knowledge base.