Diffusion LLM with Native Variable Generation Lengths: Let [EOS] Lead the Way

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Yang, Yicun, Wang, Cong, Wang, Shaobo, Wen, Zichen, Qi, Biqing, Xu, Hanlin, Zhang, Linfeng
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866908617312043008
author Yang, Yicun
Wang, Cong
Wang, Shaobo
Wen, Zichen
Qi, Biqing
Xu, Hanlin
Zhang, Linfeng
author_facet Yang, Yicun
Wang, Cong
Wang, Shaobo
Wen, Zichen
Qi, Biqing
Xu, Hanlin
Zhang, Linfeng
contents Diffusion-based large language models (dLLMs) have exhibited substantial potential for parallel text generation, which may enable more efficient generation compared to autoregressive models. However, current dLLMs suffer from fixed generation lengths, which indicates the generation lengths of dLLMs have to be determined before decoding as a hyper-parameter, leading to issues in efficiency and flexibility. To solve these problems, in this work, we propose to train a diffusion LLM with native variable generation lengths, abbreviated as dLLM-Var. Concretely, we aim to train a model to accurately predict the [EOS] token in the generated text, which makes a dLLM be able to natively infer in a block diffusion manner, while still maintaining the ability of global bi-directional (full) attention and high parallelism. Experiments on standard benchmarks demonstrate that our method achieves a 30.1x speedup over traditional dLLM inference paradigms and a 2.4x speedup relative to autoregressive models such as Qwen and Llama. Our method achieves higher accuracy and faster inference, elevating dLLMs beyond mere academic novelty and supporting their practical use in real-world applications. Codes and models have been released.
format Preprint
id arxiv_https___arxiv_org_abs_2510_24605
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Diffusion LLM with Native Variable Generation Lengths: Let [EOS] Lead the Way
Yang, Yicun
Wang, Cong
Wang, Shaobo
Wen, Zichen
Qi, Biqing
Xu, Hanlin
Zhang, Linfeng
Computation and Language
Diffusion-based large language models (dLLMs) have exhibited substantial potential for parallel text generation, which may enable more efficient generation compared to autoregressive models. However, current dLLMs suffer from fixed generation lengths, which indicates the generation lengths of dLLMs have to be determined before decoding as a hyper-parameter, leading to issues in efficiency and flexibility. To solve these problems, in this work, we propose to train a diffusion LLM with native variable generation lengths, abbreviated as dLLM-Var. Concretely, we aim to train a model to accurately predict the [EOS] token in the generated text, which makes a dLLM be able to natively infer in a block diffusion manner, while still maintaining the ability of global bi-directional (full) attention and high parallelism. Experiments on standard benchmarks demonstrate that our method achieves a 30.1x speedup over traditional dLLM inference paradigms and a 2.4x speedup relative to autoregressive models such as Qwen and Llama. Our method achieves higher accuracy and faster inference, elevating dLLMs beyond mere academic novelty and supporting their practical use in real-world applications. Codes and models have been released.
title Diffusion LLM with Native Variable Generation Lengths: Let [EOS] Lead the Way
topic Computation and Language
url https://arxiv.org/abs/2510.24605