https://txtmix.com/tags/transformerlens/Recent content in TransformerLens on Text MatrixHugozh-cnTue, 21 Jul 2026 20:06:14 +0800https://txtmix.com/posts/tech/transformerlens-skill-modular-mechanistic-interpretability/Sun, 17 May 2026 12:04:00 +0800https://txtmix.com/posts/tech/transformerlens-skill-modular-mechanistic-interpretability/<h2 id="项目概览">项目概览</h2> <p><strong>transformerlens-skill</strong>（<a href="https://github.com/Durararananke/transformerlens_skill" target="_blank" rel="noopener noreffer ">Durararananke/transformerlens_skill</a>）是一个基于 TransformerLens 框架的机制可解释性（Mechanistic Interpretability）工具库。它把激活缓存、因果追踪、归因修补（Attribution Patching）、对数透镜（Logit Lens）和激活 Steering 等常见分析范式封装成了独立模块，研究者可以直接调用而不需要每次从零写 boilerplate。</p>