阿斯塔纳时间
AI QAZAQSTANKZ · AI测试版

哈萨克斯坦人工智能与高科技领域的智能门户

Research1 分钟阅读

内容来自原始来源:

arXiv — cs.AI

在 arXiv — cs.AI 阅读全文

本内容来自公开来源。若发布内容影响您的权利 → 申请删除信息

标题为英语LatentQuant: Preserving the Policy-Facing Latent Contract under NVFP4 VAE Quantization

Recent world action models (WAMs) reuse pretrained video VAEs whose encoder latents directly condition downstream action policies. Quantization must therefore preserve not only reconstruction fidelity but also the policy-facing latent contract expected by the frozen policy.…

arXiv — cs.AI作者: Ziye Deng, Lufang Chen, Shuyu Feng, Zhenwei Duan, Zicong Ye, Yu Sun, Xiaofan Li, Ruyi Gan, Hao Wang, Hao Zhang2026年10月6日 (10月6日)
提供 1 种语言版本

AI 摘要:要点

  • Recent world action models (WAMs) reuse pretrained video VAEs whose encoder latents directly condition downstream action policies
  • Quantization must therefore preserve not only reconstruction fidelity but also the policy-facing latent contract expected by the frozen policy.…
Research
KZ·AI 3E2D
Инсайт дня

LatentQuant: Preserving the Policy-Facing Latent Contract under NVFP4 VAE Quantization

标签:#arXiv
Поделиться:

相关文章

返回列表