<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
  <title>MATRIX · RFD4-Proteina</title>
  <link>https://derekwang2002.github.io/ai-intelligence/projects/rfd4-proteina/</link>
  <description>AI 模型、Agent、工程与基础设施的重要变化，附一手来源与建议等级。</description>
  <language>zh-CN</language>
  <lastBuildDate>Mon, 02 Nov 2026 16:56:00 GMT</lastBuildDate>
  <atom:link href="https://derekwang2002.github.io/ai-intelligence/feeds/projects/rfd4-proteina.xml" rel="self" type="application/rss+xml"/>
  <item>
    <title>NVIDIA 发布 RFD4-Proteina 代码与权重</title>
    <link>https://derekwang2002.github.io/ai-intelligence/events/ev-20261021-06/</link>
    <guid isPermaLink="false">ev-20261021-06-published</guid>
    <pubDate>Wed, 21 Oct 2026 15:58:23 GMT</pubDate>
    <category>新事件</category>
    <category>TRIAL</category>
    <description>&lt;p&gt;NVIDIA 与 RosettaCommons 以 BSD-3-Clause 发布 RFD4-Proteina 代码和免申请权重。约 4.35 亿参数的模型联合设计蛋白质序列与全原子结构，reward-guided inference-time search 会从得分更高的中间候选继续分支。公开产物可以复现，但质量与 1.37–1.73 倍 BioIR 加速仍是发布方结果。&lt;/p&gt;&lt;p&gt;&lt;strong&gt;为什么读：&lt;/strong&gt;RFD4-Proteina 公开代码、权重与测试数据，让蛋白质联合设计方法可以立即复现。&lt;/p&gt;&lt;p&gt;建议等级：TRIAL&lt;/p&gt;&lt;p&gt;来源：&lt;a href=&quot;https://developer.nvidia.com/blog/design-proteins-with-rfd4-proteina-using-reward-guided-inference-time-search/&quot;&gt;RFD4-Proteina technical release&lt;/a&gt; · &lt;a href=&quot;https://github.com/RosettaCommons/RFD4-Proteina&quot;&gt;RFD4-Proteina repository&lt;/a&gt; · &lt;a href=&quot;https://huggingface.co/nvidia/RFD4-Proteina&quot;&gt;RFD4-Proteina weights&lt;/a&gt;&lt;/p&gt;</description>
  </item>
</channel>
</rss>
