Scholar Feed

Speculative decoding

ParallelSpec

Superseded baseline#42 of 151 most-superseded

Cited as a baseline — critiqued by newer work, not yet beaten on a benchmark here

3 papers critique it · 0 beat it on benchmarks

What papers say

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

Once alternatives from different depths are combined into a draft tree, they form a large combinatorial space in which many paths are not coherent continuations, and the verifier wastes budget on them.
— SpecBlock: Block-Iterative Speculative Decoding with Dynamic Tree Drafting
“these approaches still use target model information, making them inherently target-dependent”
— PARD: Accelerating LLM Inference with Low-Cost PARallel Draft Model Adaptation
“ParallelSpec~xiao2024parallelspec proposed parallel drafting with a single transformer layer, but omits critical implementation details---notably whether and how target model hidden states are utilized---and does not address the memory scaling challenges that arise from extended training sequences with multiple parallel prediction positions.”
— P-EAGLE: Parallel-Drafting EAGLE with Scalable Training

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Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.