Comment by houjing wei (houjing-wei) on Are SAE features from the Base Model still meaningful to LLaVA? · 2025-02-06T17:25:47.868Z · LW · GW

Thank you very much for this highly insightful blog! I am a beginner in SAE but am very interested in how SAE can be used to interpret modern LLM-based vision-language models.

 

May I ask if the experiments in your blog have code available for sharing?