PyPI page
Home page
Author:
Bora Kargı
License:
MIT
Summary:
PrefScope: an interpretability framework for diagnosing post-training data and model behavior with sparse autoencoders.
Latest version:
0.1.0
Required dependencies:
numpy
|
openai
|
pandas
|
pyarrow
|
pyyaml
|
scikit-learn
|
scipy
|
tqdm
|
transformers
Optional dependencies:
datasets
|
igraph
|
leidenalg
|
matplotlib
|
plotly
|
pytorch-triton-rocm
|
streamlit
|
torch
|
vllm
Downloads last day:
0
Downloads last week:
2
Downloads last month:
10
7-day smoothing