Is Block diffusion superseded?

Block diffusion (KV-cache compression): cited as a baseline — critiqued by newer work, not yet beaten on a benchmark here. 1 paper(s) critique it, 0 beat it on benchmarks — #124 of 234 most-superseded. Sub-problem: cluster led by Fast-dLLM. Newer alternatives in the same sub-problem include WaveFilter.

KV-cache compression

Block diffusion

Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models

Superseded baseline#124 of 234 most-superseded · first seen Mar 12, 2025

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

1 papers critique it · 0 beat it on benchmarks

What papers say

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

This requires considering the KV-Cache in training, making twice the forward computation in the training and its form is still constrained in the autoregressive formula.

— dKV-Cache: The Cache for Diffusion Language 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Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.

WaveFilterWaveFilter: Enhancing the Long-Context Capability of Diffusion LLMs via Wavelet-Guided KV Cache Filtering
May 30, 2026