Skip to content
Hot storiesContinuously updated

研究团队提出Q64候选评分系统

1 reports1 reporting sourcesUpdated 22 hours ago

Start here

AI overview

2026年11月4日,研究人员在arXiv发表论文,提出名为Q64的候选评分系统。该系统基于冻结的64层Qwen3.8-27B模型复用文档表示,仅通过更新88.66M参数来训练残差适配器与4层解码器。测试结果显示,在ARC-Challenge和MMLU-Pro等四项基准评估中,联合适配方案成功将宏准确率从66.46%提升至74.32%。

AI-generated from reports · Updated 22 hours ago

Report timeline

Follow the reports to see different sides of the story.

Nov 4
  1. arXiv 自然语言处理
    Q64:基于联合适配读出网络的共享上下文候选评分系统

    研究人员提出候选评分系统 Q64,基于冻结的 64 层 Qwen3.8-27B 复用文档表示,仅更新 88.66M 参数训练残差适配器与 4 层解码器。在 ARC-Challenge 和 MMLU-Pro 等四项基准评估中,联合适配将宏准确率从 66.46% 提升至 74.32%。

Story heat trend

Current heat 5·Comparable peak 10(Nov 4 14:00)·Comparable change, last 24 hours–

02.557.510Nov4Nov4Nov5Nov5

Trends compare only the same parties observed continuously, so the scope can be narrower than current heat counts. Move the pointer or tap the chart to see hourly heat; use the left and right arrow keys on a keyboard.