Product
TransformerLens
A PyTorch-based Transformer interpretability library that supports intervention and analysis of model internal activations and attention.
Timeline (last 90 days)
Aug 27
机制可解释性领域的高质量学习资源包括 TransformerLens 库、ARENA 训练营课程以及大量 arXiv 论文和博客
Unverified50%
Aug 27
TransformerLens、Baukit等研究工具库大量依赖hook来实现激活缓存、因果干预和激活修补等可解释性分析技术
Unverified50%