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
Os modelos de linguagem autoregressiva emitem um token por passagem avançada. Isto causa alta latência de inferência, especialmente quando os modelos são escalados para qualidade. Neste artigo, introduzimos modelos de linguagem paralela (PLMs), que geram tokens mutuamente consistentes em uma passagem avançada, roteando dependências através de variáveis auxiliares latentes. Os PLMs são treinados usando um limite de variação que reduz a uma perda autoregressiva mais uma perda de decodificação paralela. Empiricamente, os PLMs estão em uma fronteira de Pareto mais rápida e melhor do que as linhas de base de difusão autoregressiva e discreta. Os PLMs pré-treinados alcançam perda pré-autoregressiva perto de escalas de modelo de 80M a parâmetros 3B, e o
Machine translation (Meta M2M100); the English original follows.
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
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.26693, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange