Guardado en:
Detalles Bibliográficos
Autores principales: Wu, Junde, Hu, Minhao, Zhu, Jiayuan, Liu, Yuyuan, Zhang, Tianyi, Li, Kang, Chen, Jingkun, Pan, Jiazhen, Xu, Min, Jin, Yueming
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2603.06617
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866908870630178816
author Wu, Junde
Hu, Minhao
Zhu, Jiayuan
Liu, Yuyuan
Zhang, Tianyi
Li, Kang
Chen, Jingkun
Pan, Jiazhen
Xu, Min
Jin, Yueming
author_facet Wu, Junde
Hu, Minhao
Zhu, Jiayuan
Liu, Yuyuan
Zhang, Tianyi
Li, Kang
Chen, Jingkun
Pan, Jiazhen
Xu, Min
Jin, Yueming
contents We introduce \textbf{Evo}, a duality latent trajectory model that bridges autoregressive (AR) and diffusion-based language generation within a continuous evolutionary generative framework. Rather than treating AR decoding and diffusion generation as separate paradigms, Evo reconceptualizes text generation as a latent flow: each token is associated with a vector-valued embedding that evolves over a progression variable $t_i \in [0, 1]$, indicating its semantic maturity. Low $t_i$ values correspond to confident AR-like refinement, while high values invoke diffusion-style planning, allowing the model to adaptively balance AR and diffusion based on uncertainty. Theoretically, we show that both AR and diffusion models emerge as discretizations of a shared probability flow, and we derive Evo's training objective from a unified variational ELBO. The model is implemented as a time-conditioned Transformer governed by a shared vector field, trained end-to-end to jointly infer latent codes and their progression times. During decoding, Evo performs efficient, semantics-aware refinement, achieving high-quality outputs without sacrificing speed. Empirically, Evo 8B achieves state-of-the-art or highly competitive results on 15 diverse benchmarks, including reasoning (GSM8K, ARC-C), code generation (HumanEval, MBPP), and general language understanding, while maintaining fast inference speed. Our results demonstrate that Evo delivers a new paradigm for LLM design with strong generation quality, robust symbolic reasoning, and decoding efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2603_06617
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Evo: Autoregressive-Diffusion Large Language Models with Evolving Balance
Wu, Junde
Hu, Minhao
Zhu, Jiayuan
Liu, Yuyuan
Zhang, Tianyi
Li, Kang
Chen, Jingkun
Pan, Jiazhen
Xu, Min
Jin, Yueming
Machine Learning
Artificial Intelligence
We introduce \textbf{Evo}, a duality latent trajectory model that bridges autoregressive (AR) and diffusion-based language generation within a continuous evolutionary generative framework. Rather than treating AR decoding and diffusion generation as separate paradigms, Evo reconceptualizes text generation as a latent flow: each token is associated with a vector-valued embedding that evolves over a progression variable $t_i \in [0, 1]$, indicating its semantic maturity. Low $t_i$ values correspond to confident AR-like refinement, while high values invoke diffusion-style planning, allowing the model to adaptively balance AR and diffusion based on uncertainty. Theoretically, we show that both AR and diffusion models emerge as discretizations of a shared probability flow, and we derive Evo's training objective from a unified variational ELBO. The model is implemented as a time-conditioned Transformer governed by a shared vector field, trained end-to-end to jointly infer latent codes and their progression times. During decoding, Evo performs efficient, semantics-aware refinement, achieving high-quality outputs without sacrificing speed. Empirically, Evo 8B achieves state-of-the-art or highly competitive results on 15 diverse benchmarks, including reasoning (GSM8K, ARC-C), code generation (HumanEval, MBPP), and general language understanding, while maintaining fast inference speed. Our results demonstrate that Evo delivers a new paradigm for LLM design with strong generation quality, robust symbolic reasoning, and decoding efficiency.
title Evo: Autoregressive-Diffusion Large Language Models with Evolving Balance
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2603.06617