Automatically Identifying Local and Global Circuits with Linear Computation Graphs
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916330985226240 |
|---|---|
| author | Ge, Xuyang Zhu, Fukang Shu, Wentao Wang, Junxuan He, Zhengfu Qiu, Xipeng |
| author_facet | Ge, Xuyang Zhu, Fukang Shu, Wentao Wang, Junxuan He, Zhengfu Qiu, Xipeng |
| contents | Circuit analysis of any certain model behavior is a central task in mechanistic interpretability. We introduce our circuit discovery pipeline with Sparse Autoencoders (SAEs) and a variant called Transcoders. With these two modules inserted into the model, the model's computation graph with respect to OV and MLP circuits becomes strictly linear. Our methods do not require linear approximation to compute the causal effect of each node. This fine-grained graph identifies both end-to-end and local circuits accounting for either logits or intermediate features. We can scalably apply this pipeline with a technique called Hierarchical Attribution. We analyze three kinds of circuits in GPT-2 Small: bracket, induction, and Indirect Object Identification circuits. Our results reveal new findings underlying existing discoveries. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2405_13868 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Automatically Identifying Local and Global Circuits with Linear Computation Graphs Ge, Xuyang Zhu, Fukang Shu, Wentao Wang, Junxuan He, Zhengfu Qiu, Xipeng Machine Learning Computation and Language Circuit analysis of any certain model behavior is a central task in mechanistic interpretability. We introduce our circuit discovery pipeline with Sparse Autoencoders (SAEs) and a variant called Transcoders. With these two modules inserted into the model, the model's computation graph with respect to OV and MLP circuits becomes strictly linear. Our methods do not require linear approximation to compute the causal effect of each node. This fine-grained graph identifies both end-to-end and local circuits accounting for either logits or intermediate features. We can scalably apply this pipeline with a technique called Hierarchical Attribution. We analyze three kinds of circuits in GPT-2 Small: bracket, induction, and Indirect Object Identification circuits. Our results reveal new findings underlying existing discoveries. |
| title | Automatically Identifying Local and Global Circuits with Linear Computation Graphs |
| topic | Machine Learning Computation and Language |
| url | https://arxiv.org/abs/2405.13868 |