Discovering Millions of Interpretable Features with Sparse Autoencoders

• URL: http://arxiv.org/abs/2606.26620v1
• Date: Thu, 25 Jun 2026 05:33:03 GMT
• Title: Discovering Millions of Interpretable Features with Sparse Autoencoders
• Abstract summary: We introduce textbfQwen3-Instruct SAE, a comprehensive suite of SAEs trained on the Qwen3 instruction-tuned model family.<n>We show that selected SAE features can causally steer instruction-tuned Qwen3 models toward refusal behavior.
• Score: 7.920281725575481
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Sparse autoencoders (SAEs) have emerged as a powerful tool for decomposing superposed language model representations into sparse and interpretable features. However, training SAEs is computationally expensive, and available open-source SAE models remain limited. In this work, we introduce \textbf{Qwen3-Instruct SAE}, a comprehensive suite of SAEs trained on the Qwen3 instruction-tuned model family, covering Qwen3-1.7B, Qwen3-4B, and Qwen3-8B. For Qwen3-1.7B and Qwen3-4B, we train layer-wise SAEs at three key activation sites: residual streams, MLP outputs, and attention outputs. For Qwen3-8B, we train SAEs on a subset of residual stream layers. We systematically evaluate these SAEs using both activation-level reconstruction metrics and model-level recovery metrics, revealing distinct sparsity--fidelity trade-offs across layers and components. Finally, we demonstrate the utility of Qwen3-Instruct SAE through a refusal-steering case study, showing that selected SAE features can causally steer instruction-tuned Qwen3 models toward refusal behavior. Our release provides a practical resource for studying sparse representations, feature-level mechanisms, and behavioral interventions in instruction-tuned language models

Related papers

• Steering Vision-Language Models with Joint Sparse Autoencoders [2.367986397868328]
  We introduce the Joint Sparse Autoencoder (JSAE) to factorize sequence-pooled vision and language activations into shared, interpretable image/caption-level features.<n>Applying JSAE to vision-language models (VLMs) yields representations that are difficult to use as controllable cross-modal steering directions.
  arXiv  Detail & Related papers  (2026-06-24T10:11:04Z)
• From Sparse Features to Trustworthy Proxies: Certifying SAE-Based Interpretability [53.681678236115836]
  We study when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM.<n>Our framework derives an upper bound on the base model's expected risk using four measurable quantities.<n>We show that the bound becomes non-vacuous on GPT-2 Small, Gemma-2B, and Llama-3-8B at practical sample sizes.
  arXiv  Detail & Related papers  (2026-06-16T18:28:23Z)
• Qwen-Scope: Turning Sparse Features into Development Tools for Large Language Models [80.45129499188461]
  We introduce Qwen-Scope, an open-source suite of sparse autoencoders (SAEs) built on the Qwen model family.<n>We show that SAEs can go beyond post-hoc analysis to serve as practical interfaces for model development along four directions.
  arXiv  Detail & Related papers  (2026-05-12T10:01:06Z)
• Control Reinforcement Learning: Interpretable Token-Level Steering of LLMs via Sparse Autoencoder Features [1.5874067490843806]
  Control Reinforcement Learning trains a policy to select SAE features for steering at each token, producing interpretable intervention logs.<n> Adaptive Feature Masking encourages diverse feature discovery while preserving singlefeature interpretability.<n>On Gemma 2 2B across MMLU, BBQ, GSM8K, HarmBench, and XSTest, CRL achieves improvements while providing per-token intervention logs.
  arXiv  Detail & Related papers  (2026-02-11T02:28:49Z)
• SQS: Enhancing Sparse Perception Models via Query-based Splatting in Autonomous Driving [56.198745862311824]
  We introduce SQS, a novel query-based splatting pre-training for sparse Perception Models (SPMs)<n> SQS predicts 3D Gaussian representations from sparse queries during pre-training, leveraging self-supervised splatting to learn fine-grained contextual features.<n>Experiments on autonomous driving benchmarks demonstrate that SQS delivers considerable performance gains across multiple query-based 3D perception tasks.
  arXiv  Detail & Related papers  (2025-09-20T09:25:19Z)
• Beyond I'm Sorry, I Can't: Dissecting Large Language Model Refusal [40.3000670795883]
  Refusal on harmful prompts is a key safety behaviour in instruction-tuned large language models (LLMs)<n>We study two public instruction-tuned models, Gemma-2-2B-IT and LLaMA-3.1-8B-IT, using sparse autoencoders (SAEs) trained on residual-stream activations.<n>Given a harmful prompt, we search the SAE latent space for feature sets whose ablation flips the model from refusal to compliance, demonstrating causal influence and creating a jailbreak.
  arXiv  Detail & Related papers  (2025-09-07T02:29:07Z)
• Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models [90.54780244175511]
  We introduce the Qwen3 Embedding series, a significant advancement over its predecessor, the GTE-Qwen series.<n>The Qwen3 Embedding series offers a spectrum of model sizes for both embedding and reranking tasks.<n>The Qwen3 Embedding series achieves state-of-the-art results across diverse benchmarks.
  arXiv  Detail & Related papers  (2025-06-05T15:49:48Z)
• Ensembling Sparse Autoencoders [10.81463830315253]
  Sparse autoencoders (SAEs) are used to decompose neural network activations into human-interpretable features.<n>We propose to ensemble multiple SAEs through naive bagging and boosting.<n>Our empirical results demonstrate that ensembling SAEs can improve the reconstruction of language model activations, diversity of features, and SAE stability.
  arXiv  Detail & Related papers  (2025-05-21T23:31:21Z)
• SPaR: Self-Play with Tree-Search Refinement to Improve Instruction-Following in Large Language Models [88.29990536278167]
  We introduce SPaR, a self-play framework integrating tree-search self-refinement to yield valid and comparable preference pairs.<n>Our experiments show that a LLaMA3-8B model, trained over three iterations guided by SPaR, surpasses GPT-4-Turbo on the IFEval benchmark without losing general capabilities.
  arXiv  Detail & Related papers  (2024-12-16T09:47:43Z)
• Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders [115.34050914216665]
  Sparse Autoencoders (SAEs) have emerged as a powerful unsupervised method for extracting sparse representations from language models. 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. We assess the generalizability of SAEs trained on base models to longer contexts and fine-tuned models.
  arXiv  Detail & Related papers  (2024-10-27T17:33:49Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.