# RankE：离散文本到图像生成的端到端后训练与解码器协同进化

- 来源：HuggingFace Daily Papers（社区热门论文）
- 发布时间：2026-05-20 08:00
- AIHOT 分数：55
- AIHOT 链接：https://aihot.virxact.com/items/cmpkp4pvu08y0sl01rq91ot5g
- 原文链接：https://arxiv.org/abs/2605.21195

## AI 摘要

当前离散自回归文本到图像模型的后训练通常只优化策略网络而固定VQ解码器，导致潜在协变量偏移，使奖励提升但图像质量下降。为此，本文提出首个端到端后训练框架RankE，通过交替优化策略与解码器实现协同进化。在LlamaGen-XL（775M）上，标准RL仅提升CLIP但恶化FID，而RankE同时改善两者（MS-COCO 30K上FID 15.21, CLIP 33.76）。在Janus-Pro（1B）上验证了其稳定转化奖励为图像质量的能力。

## 正文

Discrete autoregressive (AR) text-to-image (T2I) models pair a VQ tokenizer with an AR policy, and current post-training pipelines optimize only the policy while keeping the VQ decoder frozen. Recent diffusion T2I work, exemplified by REPA-E, has shown that the VAE itself constitutes a key alignment bottleneck, yet no analogous investigation exists for discrete AR models. We show that policy-only optimization induces Latent Covariate Shift: as the policy evolves, the resulting token distribution diverges from the ground-truth distribution on which the decoder was trained, such that reward scores improve while decoded image quality degrades. To address this mismatch, we propose RankE, the first end-to-end post-training framework for discrete T2I generation. Rather than optimizing the policy against a fixed decoder, RankE co-evolves both components through alternating optimization: each module maximizes a ranking-based alignment objective while being regularized by a stability-preserving anchor suited to its parameter space. This co-evolution breaks the fidelity--alignment trade-off that plagues frozen-decoder approaches: on LlamaGen-XL (775M), standard RL improves CLIP but degrades FID, whereas RankE improves both simultaneously (FID 15.21, CLIP 33.76 on MS-COCO 30K). Consistent gains on Janus-Pro (1B) confirm that decoder co-evolution reliably converts reward optimization into pixel-space quality improvements.