https://jaehun.me/en/tags/flow-matching/Recent content in Flow Matching on Jaehun's BlogHugoen-USSat, 24 Oct 2026 08:16:46 +0900https://jaehun.me/en/posts/paper-2610-28559v1/Sat, 24 Oct 2026 00:00:00 +0900https://jaehun.me/en/posts/paper-2610-28559v1/<p><a href="https://arxiv.org/abs/2610.28559v1"target="_blank" class="inline-flex items-center gap-1" >Paper<svg class="h-3 w-3 flex-shrink-0" id="external-link" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M15 3h6v6m-11 5L21 3m-3 10v6a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2V8a2 2 0 0 1 2-2h6"/></svg> </a></p> <h2 id="the-sequential-bottleneck-of-sampling-one-token-at-a-time">The sequential bottleneck of sampling one token at a time<a href="#the-sequential-bottleneck-of-sampling-one-token-at-a-time" class="heading-anchor" aria-label="Link to this section"></a></h2><p>Large language models can compute an entire sentence in parallel during training, but when generating text they must know the preceding token before they can choose the next one. No matter how fast the GPU is, this order cannot be skipped in one pass, so generation latency grows. This post covers Lafa (Large flow language model), which attacks this bottleneck with block-level flow.</p>