atp_star

KO | EN

gitlite — search

#javascript #react #ai #typescript #llm #go #golang #rust #deep-learning #docker #cli #ios #openai #open-source

(★ 20)

PyTorch and NNsight implementation of AtP* (Kramar et al 2024, DeepMind)

atp_star 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# AtP*

### Improved Attribution Patching for Localizing Large Model Behaviour

This repo contains code to perform the AtP* algorithm for improved Attribution Patching. 
The code is based on the [AtP*: An efficient and scalable method for localizing LLM behaviour to components](https://arxiv.org/pdf/2403.00745.pdf), Kramar et al. 2024 from DeepMind.
This codebase is an unofficial replication and any errors are due to myself rather than the paper authors.

**Attribution Patching** (AtP) was introduced in [Nanda 2022](https://www.neelnanda.io/mechanistic-interpretability/attribution-patching) as a quick approximation to the more precise _Activation Patching_ (AcP) which details the contribution of each component to some metric (e.g. NLL loss, IOI score, etc.). It works by taking the first order Taylor approximation of the contribution c(n).

## Appreciation

Thanks to Jaden and the nnsight team for the `nnsight` package that is used for the caching and interventions.

Thanks to Alice and the MechInterp Discord for the discussions and feedback.

## Contributions

Contributions are welcome, please feel free to raise PRs to implement additional features you're interested in! 😄

## Progress

- [x] Implement AtP algorithm
- [x] Implement AtP with QK-Fix algorithm improvements
- [x] Implement full AtP* with GradDrop
- [x] Look at MLP component contributions
- [ ] Optimise for GPU and apply fast K_fix method (Algorithm 4)
- [ ] Conduct ablations and throughput experiments to reproduce paper results
- [ ] Testing
- [ ] Decouple from GPT-2
- [ ] Add complete circuit-finding algorithm with subsampling and sending the highest ranked nodes to the slower AcP algorithm
- [ ] Add subsampling for diagnostic bounds

atp_star

atp_star 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

atp_star

atp_star 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README