Are Single-Token Sparse Autoencoder Features Causally Necessary? Layer-Depth and SAE-Family Effects
cs.LG, cs.CL
Submitted: 2026-07-22
Updated: 2026-08-29
Comments: 23 pages, 10 figures, 27 tables
Project page: https://skylion007.github.io/OpenWebTextCorpus
License: http://creativecommons.org/licenses/by/4.0/
The gist: Sparse autoencoder (SAE) features are used to interpret and steer large language models, yet nobody has tested whether a feature's causal role is stable across SAE families.
Terminology
Abstract
Sparse autoencoder (SAE) features are used to interpret and steer large language models, yet nobody has tested whether a feature's causal role is stable across SAE families. Single-token features fire on one vocabulary item, so ground truth permits direct comparison. We analyze 3.9M features across six models and three SAE families and zero-ablate at full layer depth: they sit 4.7x tighter in decoder space and concentrate in early layers. Deleting one lowers the model's logit for that token in 178 of 208 layer conditions, significant after multiple-comparison correction. And depth decides how the damage lands: early-layer deletions disrupt the layers that follow, late-layer deletions change the output directly. Cross-family causal differences exceed within-family scale effects: on the same base model, GemmaScope and BatchTopK features are causally anchored, LlamaScope features locally redundant. Under LlamaScope the token returns to within 2x its pre-ablation rank 96-98% of the time. Changing only the activation function reverses the sign of that difference, so the training recipe is the remaining candidate: cross-family claims are sensitive to training methodology, not just activation function or scale.
Sources
- Evolution of SAE Features Across Layers in LLMs
- The Llama 3 Herd of Models
- Improving Steering Vectors by Targeting Sparse Autoencoder Features
- A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoders
- Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders
- Projecting Assumptions: The Duality Between Sparse Autoencoders and Concept Geometry
- Not All Language Model Features Are One-Dimensionally Linear
- Sanity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines?
- Quantifying Feature Space Universality Across Large Language Models via Sparse Autoencoders
- Gemma 3 Technical Report
- The Geometry of Concepts: Sparse Autoencoder Feature Structure
- Sparse Autoencoders Trained on the Same Data Learn Different Features
- Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders
- A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Models
- Gemma 2: Improving Open Language Models at a Practical Size
- Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control
- k-Sparse Autoencoders
- SAGE: Scalable Ground Truth Evaluations for Large Sparse Autoencoders
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks