Is FastDLLM superseded?

FastDLLM (Speculative decoding): superseded — cited as a baseline and beaten by newer methods. 1 paper(s) critique it, 1 beat it on benchmarks — #53 of 151 most-superseded. Sub-problem: cluster led by SSD. Newer alternatives in the same sub-problem include FeF-DLLM, DualDiffusion, Principled Coarse-Graining (PCG).

Speculative decoding

FastDLLM

Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding

Superseded baseline#53 of 151 most-superseded · first seen May 28, 2025

Superseded — cited as a baseline and beaten by newer methods

1 papers critique it · 1 beat it on benchmarks

What papers say

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

Block-wise caching in FastDLLM fixes token representations within blocks, preventing iterative refinement.

— DualDiffusion: A Speculative Decoding Strategy for Masked Diffusion Models

Beaten on benchmarks

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

DualDiffusion beats FastDLLM
0.47 vs 0.39
Accuracy · [no_condition]
DualDiffusion: A Speculative Decoding Strategy for Masked Diffusion Models

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Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.