arrakis
arrakis copied to clipboard
yash-srivastava19
Arrakis is a library to conduct, track and visualize mechanistic interpretability experiments.
One of the ways in which we can reduce experimentation time is if we can run arrakis using YAML files. We can specify what tools we can use in this...
The docs for the project are available [here](https://arrakis-mi.readthedocs.io/en/latest/README.html). I want help in maintaining them, with resources, explanation and examples usage of tools. API reference is generated automatically, so no need...
In the future folder, I've implemented a simple CLI to use arrakis(with limited support). I want to expand it and add more support to it. PRs and discussions welcome. Read...
In the first version of Arrakis, the following tools are supported(with built in function) : [Attention Head Composition](https://arrakis-mi.readthedocs.io/en/latest/AttentionHeadComposition.html) [Attention Tools](https://arrakis-mi.readthedocs.io/en/latest/AttentionTools.html) [Causal Tracing Intervention](https://arrakis-mi.readthedocs.io/en/latest/CausalTracingIntervention.html) [Knowledge Graph Extractor](https://arrakis-mi.readthedocs.io/en/latest/KnowledgeGraphExtractor.html) [Knowledge Prober](https://arrakis-mi.readthedocs.io/en/latest/KnowledgeProber.html) [Logit Attribution](https://arrakis-mi.readthedocs.io/en/latest/LogitAttribution.html)...
Metadata
Owner
Metadata
Arrakis is a library to conduct, track and visualize mechanistic interpretability experiments.