avatar
Articles
61
Tags
115
Categories
3
Home
Archives
Tags
Categories
About
Graph
Github
The Great LibraryPaper Summary - No Long Context Back to Home
Search
Home
Archives
Tags
Categories
About
Graph
Github

Paper Summary - No Long Context

Created2026-11-10|Updated2026-11-11|Tech
|Post Views:

TODO: 正文待补充。 论文: You don’t need long context · arXiv:2611.05306 · 未查到录用信息(技术报告) · NVIDIA

Author: Askemi Wang
Link: https://wangyier.top/paper-summary-no-long-context/
Copyright Notice: All articles on this blog are licensed under CC BY-NC-SA 4.0 unless otherwise stated.
Next
Paper Summary - GTA1
This work focuses on 3 things: 1. in Agent's react loop, ask VLM rollout K candidate actions for next step, then ask Judge select most proximate one. 2. This work has trained model by GRPO with single `click reward` which outperforms other models. 3. Data cleanup: this work uses many open-source grounding training data to train the grounding model. The origin data is scraped by programs which is cheap but may contain errors. The team uses OmniParser to cleanup the origin training data.

Comments
avatar
Askemi Wang
Personal notes of everything.
Articles
61
Tags
115
Categories
3
Follow Me
Recent Posts
Paper Summary - No Long Context2026-11-10
Paper Summary - GTA12026-11-05
Paper Summary - OSWorld-G2026-11-02
Experiments2026-10-28
Paper Summary - Agent S22026-10-24
The Great Library
AboutKnowledge Graph
Browse
ArchivesTagsCategories
Elsewhere
GitHubEmail
© 2026 By Askemi Wang|Framework Hexo 8.1.2|Theme Butterfly 5.7.0
🕒 本站已运行 天 小时 分 秒
Search
Loading Database