Newest papers

Parallel Language Models via Variational Decoding

Felix Draxler, Robin C. Geyer, Daniel Geyfman, Farrin Marouf Sofian, Justus C. Will, Philipp Hildebrandt, Giosue Migliorini, Prakhar Srivastava, Gagan Sai Trinadh Duggaraju Peddiraju, Kushagra Pandey, Jorge da Silva Gonçalves, ELIJAH TAVARES, Jan Niklas Groeneveld, Yang Meng, Julia Elisabeth Vogt, Stephan Mandt · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26693 · cited by 0 works (OpenAlex) · open-access PDF

Spanish

Los modelos de lenguaje autoregresivo emiten un token por paso adelante. Esto causa una alta latencia de inferencia, en particular a medida que los modelos se escalan para la calidad. En este artículo, introducimos los modelos de lenguaje paralelo (PLMs), que generan tokens mutuamente consistentes en un paso adelante mediante el enrutamiento de dependencias a través de variables auxiliares latentes. Los PLMs se entrenan utilizando un límite de variación que se reduce a una pérdida autoregresiva más una pérdida de decodificación paralela. Empiricamente, los PLMs están en una frontera de Pareto más rápida y mejor que las líneas de base de difusión autoregresiva y discreta. Los PLMs pretrainados obtienen pérdidas pre-trenadas casi autoregresivas en escalas de

Machine translation (Meta M2M100); the English original follows.

Abstract

Autoregressive language models emit one token per forward pass. This causes high inference latency, in particular as models are scaled up for quality. In this paper, we introduce Parallel Language Models (PLMs), which generate mutually consistent tokens in one forward pass by routing dependencies through latent auxiliary variables. PLMs are trained using a variational bound that reduces to an autoregressive loss plus a parallel decoding loss. Empirically, PLMs are on a faster and better Pareto frontier than autoregressive and discrete diffusion baselines. Pretrained PLMs achieve near-autoregressive pretraining loss on model scales of 80M to 3B parameters, and speedup follows a power law in model size. When post-trained, our 1.3B model matches or exceeds an identically trained autoregressive model on math, coding, and instruction-following benchmarks, speeding up inference $2.3\text{-}5.7\times$. Together, we show that parallel generation can be trained from a single training objective at autoregressive quality. We will make our code available at https://github.com/mandt-lab/plm.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Autoregressive model, Language model, Computer science, Inference, Decoding methods, Algorithm

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange