https://jaehun.me/en/tags/2610.36671/Technical notes by Jaehun Ryu, a deep learning compiler engineer, on LLM inference, KV caches, compilers, C++, AI papers, and open-source code.en-UShttps://jaehun.me/en/posts/paper-2610-36671/https://jaehun.me/en/posts/paper-2610-36671/Sat, 31 Oct 2026 00:00:00 +0900<p><a href="https://arxiv.org/abs/2610.36671"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-problem-re-reading-the-image-at-every-step">The problem: re-reading the image at every step<a href="#the-problem-re-reading-the-image-at-every-step" class="heading-anchor" aria-label="Link to this section"></a></h2><p>Diffusion language models that answer questions from images do not complete a response in one pass. While filling in the blanks over many steps, they re-read an image that barely changes from the beginning at every step. LOWN asks &ldquo;does this step really need the image?&rdquo; every time, and re-reads only when necessary, cutting decoding cost substantially.</p>