Metadata-Version: 2.4
Name: inverse-cai
Version: 0.4.1
Summary: Inverse Constitutional AI (ICAI) pipeline for interpreting pairwise preferences
Author: timokau
Author-email: rdnfn <hi@arduin.io>
License-File: LICENSE
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Classifier: Programming Language :: Python :: 3
Requires-Python: >=3.10
Requires-Dist: alpaca-eval<=0.6.5
Requires-Dist: anthropic
Requires-Dist: datasets
Requires-Dist: hydra-core
Requires-Dist: langchain
Requires-Dist: langchain-anthropic
Requires-Dist: langchain-community
Requires-Dist: langchain-openai
Requires-Dist: loguru
Requires-Dist: matplotlib
Requires-Dist: pandas
Requires-Dist: python-dotenv
Requires-Dist: scikit-learn
Requires-Dist: setuptools
Provides-Extra: app
Requires-Dist: gradio; extra == 'app'
Requires-Dist: ipywidgets; extra == 'app'
Requires-Dist: jupyter; extra == 'app'
Requires-Dist: nbconvert; extra == 'app'
Requires-Dist: plotly; extra == 'app'
Provides-Extra: dev
Requires-Dist: black==24.10.0; extra == 'dev'
Requires-Dist: bump-my-version; extra == 'dev'
Requires-Dist: pytest-asyncio; extra == 'dev'
Requires-Dist: pytest>=6.0; extra == 'dev'
Description-Content-Type: text/markdown

This package contains the official implementation of the *Inverse Constitutional AI* (ICAI) algorithm [[paper]](https://arxiv.org/abs/2406.06560). ICAI compresses pairwise preference datasets into a readable list of principles (constitution) that the annotations appear to follow (e.g. "select the friendlier response"). ICAI principles provide an interpretable overview of a feedback dataset.

See the GitHub repo for docs and details: https://github.com/rdnfn/icai