Closed kernel-loophole closed 1 month ago
Hi. I updated the script with the required changes for it to run, but as the header of the script said, this script had still not updated to the most recent changes we made to the library and might not be completely functional. Is there a reason you want to start with simulation instead of using the basic.py script?
@SrGonao I've trained a sparse autoencoder and have a bunch of weights for this. I'd like to run this through the Eleuther auto-interp
The basic.py demo is the best place to start.
okay that's great
while trying to run this
https://github.com/EleutherAI/sae-auto-interp/blob/main/demo/simulate.py
script got this error ImportError: cannot import name 'FeatureLoader' from 'sae_auto_interp.features'.i also check that there is no class named FeatureLoader in features.py. checkout this colab notebook for more details https://colab.research.google.com/drive/1qX2USmeCPyFksARj0ipWVk-aYScr1uCn?usp=sharing