https://jaehun.me/en/tags/inference-efficiency/Recent content in Inference-Efficiency on Jaehun's BlogHugoen-USTue, 08 Sep 2026 10:56:16 +0000https://jaehun.me/en/posts/paper-review-continuous-autoregressive-language-models/Fri, 26 Dec 2025 00:00:00 +0900https://jaehun.me/en/posts/paper-review-continuous-autoregressive-language-models/<p><a href="https://arxiv.org/abs/2510.27688v1"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="calm-bypassing-the-token-by-token-bottleneck-with-continuous-vector-by-vector-likelihood-free-language-modeling">CALM: Bypassing the Token-by-Token Bottleneck with &ldquo;Continuous Vector-by-Vector&rdquo; Likelihood-Free Language Modeling<a href="#calm-bypassing-the-token-by-token-bottleneck-with-continuous-vector-by-vector-likelihood-free-language-modeling" class="heading-anchor" aria-label="Link to this section"><svg class="h-4 w-4" aria-hidden="true" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><g fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2"><path d="M10 13a5 5 0 0 0 7.54.54l3-3a5 5 0 0 0-7.07-7.07l-1.72 1.71"/><path d="M14 11a5 5 0 0 0-7.54-.54l-3 3a5 5 0 0 0 7.07 7.07l1.71-1.71"/></g></svg></a></h2><p><strong>CALM</strong> (Continuous Autoregressive Language Models) compresses tokens into continuous latent vectors at a rate of (K=4) (tokens/step) and then autoregressively generates the next vector, claiming a &ldquo;performance–compute frontier shift&rdquo; that lowers training FLOPs and inference FLOPs/token at once at <strong>comparable quality (BrierLM)</strong>. (source: §7.2, Tab.1)</p>