04:00
2026-09-14
arxiv.org
large-language-models
Fixed State, Long Reach: What a Constant-Size Cache Buys Block Diffusion at Scale
A study on arXiv (2609.11998v1) pretrained three 3B block-diffusion denoisers — attention, Mamba, and hybrid — on 300B tokens and found that only the state-space cache stays O(1) in sequence length, d…