Evolution of SAE Features Across Layers in LLMs
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_ | 1866915024161734656 |
|---|---|
| author | Balcells, Daniel Lerner, Benjamin Oesterle, Michael Ucar, Ediz Heimersheim, Stefan |
| author_facet | Balcells, Daniel Lerner, Benjamin Oesterle, Michael Ucar, Ediz Heimersheim, Stefan |
| contents | Sparse Autoencoders for transformer-based language models are typically defined independently per layer. In this work we analyze statistical relationships between features in adjacent layers to understand how features evolve through a forward pass. We provide a graph visualization interface for features and their most similar next-layer neighbors (https://stefanhex.com/spar-2024/feature-browser/), and build communities of related features across layers. We find that a considerable amount of features are passed through from a previous layer, some features can be expressed as quasi-boolean combinations of previous features, and some features become more specialized in later layers. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_08869 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Evolution of SAE Features Across Layers in LLMs Balcells, Daniel Lerner, Benjamin Oesterle, Michael Ucar, Ediz Heimersheim, Stefan Machine Learning Sparse Autoencoders for transformer-based language models are typically defined independently per layer. In this work we analyze statistical relationships between features in adjacent layers to understand how features evolve through a forward pass. We provide a graph visualization interface for features and their most similar next-layer neighbors (https://stefanhex.com/spar-2024/feature-browser/), and build communities of related features across layers. We find that a considerable amount of features are passed through from a previous layer, some features can be expressed as quasi-boolean combinations of previous features, and some features become more specialized in later layers. |
| title | Evolution of SAE Features Across Layers in LLMs |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2410.08869 |