REVIEW 29 cited by
Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders
read the original abstract
Sparse Autoencoders (SAEs) have emerged as a powerful unsupervised method for extracting sparse representations from language models, yet scalable training remains a significant challenge. We introduce a suite of 256 SAEs, trained on each layer and sublayer of the Llama-3.1-8B-Base model, with 32K and 128K features. Modifications to a state-of-the-art SAE variant, Top-K SAEs, are evaluated across multiple dimensions. In particular, we assess the generalizability of SAEs trained on base models to longer contexts and fine-tuned models. Additionally, we analyze the geometry of learned SAE latents, confirming that \emph{feature splitting} enables the discovery of new features. The Llama Scope SAE checkpoints are publicly available at~\url{https://huggingface.co/fnlp/Llama-Scope}, alongside our scalable training, interpretation, and visualization tools at \url{https://github.com/OpenMOSS/Language-Model-SAEs}. These contributions aim to advance the open-source Sparse Autoencoder ecosystem and support mechanistic interpretability research by reducing the need for redundant SAE training.
Forward citations
Cited by 29 Pith papers
-
Descriptive Collision in Sparse Autoencoder Auto-Interpretability: When One Explanation Describes Many Features
Many distinct SAE features share identical explanations, with the average annotation resolving only 70% of feature identity in a large annotated dataset.
-
How Far Do Auto-Interpretation Labels Generalize: A Controlled Study Across Languages, Scripts, and Rewordings
Auto-interpretation labels for SAE features generalize poorly across languages and scripts, missing the same semantic content up to 4x more often in Serbian than English and more in Cyrillic than Latin despite determi...
-
Interpretability-Guided Layer Selection over Subspace Projection: SAEs as Stethoscopes, Not Scalpels, for Raw Task Vector Model Editing
SAEs used for layer selection with raw task vectors outperform subspace projection and raise math reasoning accuracy on Gemma-3-4B-IT.
-
Sign-Aware Gated Sparse Autoencoders: Modeling Anticorrelated Features with Bi-Jump-ReLU Activations
SA-GSAE with Bi-Jump-ReLU enables one latent to encode both polarities of anticorrelated features, Pareto-dominating or matching full-width gated SAEs while reducing dead latents by up to 500x on some LLM hookpoints.
-
ReSAE: Residualized Sparse Autoencoders for Multi-Layer Transformer Interventions
ReSAEs improve multi-layer SAE interventions on Pythia-1.4B and Gemma-2-9B by training later-layer dictionaries on residuals after affine mapping, recovering more cross-entropy loss despite lower raw variance reconstruction.
-
How Language Models Process Negation
LLMs process negation using both attention-based suppression and constructive representation mechanisms (construction dominant), with late-layer attention shortcuts explaining poor accuracy on negation tasks.
-
How Language Models Process Negation
LLMs implement both attention-based suppression and constructive representations for negation, with construction dominant, despite poor accuracy from late-layer attention shortcuts.
-
Bucketing the Good Apples: A Method for Diagnosing and Improving Causal Abstraction
A four-step recipe partitions the input space using interchange intervention behavior to diagnose where causal abstractions hold and to guide improvements, demonstrated by recovering a full hypothesis from scratch in ...
-
Are LLM Uncertainty and Correctness Encoded by the Same Features? A Functional Dissociation via Sparse Autoencoders
Uncertainty and correctness in LLMs are encoded by distinct feature populations, with suppression of confounded features improving accuracy and reducing entropy.
-
When Are Sparse Feature Interventions Actually Localized? Matched Evaluation for SAE-Based Safety Control
SAE safety ablations are regime-dependent and baseline-dependent: medium-k heads can look efficient, but surface-matched dense steering often beats them and high-k collapses coherence.
-
Minimizing Targeted Activations: Input-Only Suppression of Evaluation-Awareness Latents in Large Language Models
Input-only prompt optimization suppresses an internal evaluation-awareness latent (z≈-7) and fully turns off a causally-validated SAE feature, but this does not translate to behavioral control; a placebo random direct...
-
Where Steering Signals Come From: Activation Source Selection in Activation Steering
Activation steering works best when the signal comes from the state where the model is about to produce the target behavior, not from text that already shows it.
-
Are Single-Token Sparse Autoencoder Features Causally Necessary? Layer-Depth and SAE-Family Effects
A single-token feature's causal necessity under zero-ablation depends on which SAE family found it: GemmaScope and BatchTopK features stay causally anchored while LlamaScope features are locally redundant.
-
Sign-Aware Gated Sparse Autoencoders: Modeling Anticorrelated Features with Bi-Jump-ReLU Activations
A half-width sign-aware gated sparse autoencoder matches full-width Gated SAE reconstruction on six LLM hookpoints while cutting dead features by 0.35–0.82 absolute at matched sparsity.
-
Multilingual Steering by Design: Multilingual Sparse Autoencoders and Principled Layer Selection
Multilingual SAEs strengthen cross-lingual representations for reliable steering and an intersection-based rule selects effective layers without exhaustive search.
-
Steering grids for sparse-autoencoder features: when a top-context label names an activation regime rather than a causal axis
A steering-grid protocol that varies coefficient and joint feature set shows that standard single-feature SAE labels are often incomplete, naming an activation regime rather than a causal axis.
-
Steering grids for sparse-autoencoder features: when a top-context label names an activation regime rather than a causal axis
Pairwise matrices for SAEs demonstrate that single-feature inspection mislabels causal axes, with joint suppression and matched-geometry controls revealing distinct output regimes not captured by single-feature or ran...
-
Knowledge Vector of Logical Reasoning in Large Language Models
Distinct linear knowledge vectors for deductive, inductive, and abductive reasoning in LLMs can be refined via complementary subspace constraints to improve performance through mutual knowledge sharing.
-
The Past Is Not Past: Memory-Enhanced Dynamic Reward Shaping
MEDS improves LLM RL performance by up to 4.13 pass@1 and 4.37 pass@128 points by dynamically penalizing rollouts matching prevalent historical error clusters identified via memory-stored representations and density c...
-
LangFIR: Discovering Sparse Language-Specific Features from Monolingual Data for Language Steering
LangFIR isolates sparse language-specific SAE features via random-token filtering on monolingual data and uses them to steer multilingual LLM generation better than parallel-data methods.
-
LangFIR: Discovering Sparse Language-Specific Features from Monolingual Data for Language Steering
Language-specific sparse autoencoder features can be identified from monolingual data alone by filtering out features that also activate on random-token sequences, and steering with these features improves language control.
-
Learning Self-Interpretation from Interpretability Artifacts: Training Lightweight Adapters on Vector-Label Pairs
Training a d_model+1-parameter affine adapter on vector-label pairs lets frozen LMs label their own internal features, beating untrained self-interpretation and the noisy training labels themselves.
-
Language Model Circuits Are Sparse in the Neuron Basis
MLP neuron activations are shown to be as sparse and faithful a basis for circuit tracing as sparse autoencoder features, enabling simplified interpretability pipelines.
-
NeuroCogMap Reveals Cognitive Organization of Large Language Models
NeuroCogMap maps LLM internal representations into stable functional parcels tied to cognitive functions, failure modes, and human cortical activity during language tasks.
-
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
The survey organizes mechanistic interpretability techniques into a Locate-Steer-Improve framework to enable actionable improvements in LLM alignment, capability, and efficiency.
-
Safe-SAIL: Towards a Fine-grained Safety Landscape of Large Language Models via Sparse Autoencoder Interpretation Framework
Safe-SAIL supplies a pre-explanation metric and segment-level simulation to interpret 1758 safety SAE features across pornography, politics, violence, and terror, with public models and tools released.
-
Post-AGI Economies: Superposition and the Second Fundamental Theorem of Welfare Economics
An autonomy-qualified Second Welfare Theorem is stated for post-AGI economies under the joint conditions of convexity, stable moral status, non-fungible rights, welfare selection, non-manipulation, governed self-modif...
-
From Weights to Activations: Is Steering the Next Frontier of Adaptation?
Steering is positioned as a distinct adaptation paradigm that uses targeted activation interventions for local, reversible behavioral changes without parameter updates.
-
Discovering Millions of Interpretable Features with Sparse Autoencoders
Trains and releases SAEs for Qwen3-1.7B/4B/8B models with layer-wise coverage and demonstrates causal steering of refusal via selected features.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.