Repository navigation

gpt-viz

Public

About

A practical look at adversarial "distractors" attack described in Weight-sparse transformers have interpretable circuits (https://arxiv.org/abs/2511.13653)

0 stars

0 watching

Report repository

Name		Name	Last commit message	Last commit date
.gitignore		.gitignore
README.md		README.md
app.py		app.py
requirements.txt		requirements.txt

Repository files navigation

"Weight-sparse transformers have interpretable circuits" Implementation

Move the slider to see 'distractor' attack.

Value of 32 leads to context dilution for neuron tracking double-brackets

About

A practical look at adversarial "distractors" attack described in Weight-sparse transformers have interpretable circuits (https://arxiv.org/abs/2511.13653)

0 stars

0 watching

Report repository

Releases

Packages

Contributors

Languages