BriefGPT - AI 论文速递 ·

Significance-Driven Dynamic Token Pruning for Large-Scale Language Models

💡 原文英文，约100词，阅读约需1分钟。

📝

内容提要

本研究提出了一种显著性驱动的动态令牌剪枝框架（SDTP），旨在解决大规模语言模型在长序列推理中的计算复杂性问题。SDTP通过分层剪枝65%的输入令牌，显著降低计算量，提高推理速度，同时保持模型性能。

🏷️

OpenAI会卖给你Astra，但不会卖给你在ARC-AGI-3上得分98.6%的那套系统
文章探讨了AI模型评测中“框架工程”（harness）的关键作用。以GPT-6 Astra为例，同一模型在不同评测框架下得分差异显著（62.7%对98.6...
Airbnb Cuts Authentication Code by 60% with Server Driven Architecture
Airbnb redesigned its authentication architecture around server driven flows ...
Minmax H3中英提示词效果差异测试 - 蝈蝈俊
对于MiniMax H3，使用中文和英文提示词时，能看到都在说，英文提示词通常能带来更好的生成效果。但是我实际测试下来，感觉差别不大。我的测试是基于北京...
OpenAI admits to German wiki ‘incident’
OpenAI says it needs to overhaul how and when it reports instances of AI mode...
Robotaxis enter their villain era
It's Bullitt meets Christine meets Waymo. A new short film imagines a San...
Beyond Zero: Google Publishes Successor to BeyondCorp
In a recent research paper, Google introduced Beyond Zero, a “security model ...