Paper deep dive
Neuronpedia: Interactive SAE Feature Explorer
Johnny Lin
Models: DeepSeek-R1-Distill-Llama-8B, GPT-2 Small, GPT-OSS-20B, Gemma 2, Gemma 3, Llama 3.1, Llama 3.3, Qwen
Intelligence
Status: succeeded | Model: google/gemini-3.1-flash-lite-preview | Prompt: intel-v1 | Confidence: 95%
Last extracted: 3/11/2026, 12:34:09 AM
Summary
Neuronpedia is an open-source interpretability platform designed for exploring, steering, and experimenting with AI models. It provides access to over four terabytes of activations, explanations, and metadata, supporting tools like Sparse Autoencoders (SAEs), circuit tracing, and model steering.
Entities (6)
Relation Signals (3)
Johnny Lin → created → Neuronpedia
confidence 100% · Neuronpedia was created by Johnny Lin
Neuronpedia → supports → Sparse Autoencoder
confidence 95% · Neuronpedia supports probes, latents/features, custom vectors, concepts, and more.
Google DeepMind → developed → Gemma 3
confidence 90% · Gemma Scope 2: Comprehensive Suite of SAEs and Transcoders for Gemma 3
Cypher Suggestions (2)
Find all models supported by the Neuronpedia platform · confidence 90% · unvalidated
MATCH (p:Platform {name: 'Neuronpedia'})-[:SUPPORTS]->(m:Model) RETURN m.nameIdentify organizations that developed models hosted on Neuronpedia · confidence 85% · unvalidated
MATCH (o:Organization)-[:DEVELOPED]->(m:Model) RETURN o.name, m.name
Abstract
Open Source Interpretability Platform
Tags
Links
- Source: https://www.neuronpedia.org/
- Canonical: https://www.neuronpedia.org/
- Code: https://github.com/hijohnnylin/neuronpedia
Full Text
5,010 characters extracted from source content.
Expand or collapse full text
Get UpdatesQuarterly newsletter from our blog. No spam, unsubscribe any time.SubmitDec 2025Gemma Scope 2Google DeepMindBrowseReleaseTutorialNotebookHuggingFaceDeepMindBlogJanuary 2026Assistant AxisLu et al.Anthropic Icon Streamline Icon: https://streamlinehq.comAnthropicLaunchDemoPaperAugust 2025The Circuit AnalysisResearch LandscapeAnthropic, EleutherAI, Goodfire AI, Google DeepMindRead PostCircuitTracerWatchDemoNeuronpedia is an open source interpretability platform.Explore, steer, and experiment on AI models.GitHubGet StartedGemma ScopeExploreBrowse over four terabytes of activations, explanations, and metadata. Neuronpedia supports probes, latents/features, custom vectors, concepts, and more.ReleasesGemma Scope 2: Comprehensive Suite of SAEs and Transcoders for Gemma 3Language Model Interpretability Team, Google DeepMindTemporal Feature AnalysisLubana, Rager, Hindupur, et al.gpt-oss BatchTopK SAEsAndy ArditiFinding Misaligned Persona Features in Open-Weight ModelsAndy ArditiCircuit Tracer TranscodersHanna & PiotrowskiA Bunch of Matryoshka SAEsDavid ChaninLlama 3.3 70B Instruct SAEGoodfireLlama Scope R1: SAEs for DeepSeek-R1-Distill-Llama-8B OpenMOSS Team, Fudan UniversityGemma Scope - Exploring the Inner Workings of Gemma 2Language Model Interpretability Team, Google DeepMindAxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoderspyvene.ai, The Stanford NLP GroupLlama Scope: SAEs for Llama-3.1-8BOpenMOSS Team, Fudan UniversityFeature Splitting for GPT2-SmallJoseph BloomMulti TopK SAE for Llama3.1-8BEleutherAISparse Autoencoder for GPT2-Small - v5OpenAIIdentifying Functionally Important Features with End-to-End Sparse Dictionary LearningApollo Research · Jordan TaylorTranscoders Enable Fine-Grained Interpretable Circuit Analysis for Language ModelsJacob Dunefsky · Philippe ChlenskiSparse Autoencoders for Pythia-70M-DedupedUnder Peer ReviewAttention SAE Research PaperUnder Peer ReviewOpen Source Sparse Autoencoders for all Residual Stream Layers of GPT2-SmallJoseph BloomModelsCIRCUITGPT-PYTHONCircuitGPT-PythonOpenAIGEMMA-3-27BGemma-3-27BGoogle DeepmindGEMMA-3-12BGemma-3-12BGoogle DeepmindGEMMA-3-270M-ITGemma-3-270M-ITGoogle DeepmindGEMMA-3-1B-ITGemma-3-1B-ITGoogle DeepmindGEMMA-3-4B-ITGemma-3-4B-ITGoogle DeepmindGEMMA-3-12B-ITGemma-3-12B-ITGoogle DeepmindGEMMA-3-27B-ITGemma-3-27B-ITGoogle DeepmindGEMMA-3-270MGemma-3-270MGoogle DeepmindGEMMA-3-4BGemma-3-4BGoogle DeepmindGEMMA-3-1BGemma-3-1BGoogle DeepmindGEMMA-2-27BGemma-2-27BGoogle DeepmindGPT-OSS-20BGPT-OSS-20BOpenAIQWEN2.5-7B-ITQwen2.5-7B-ITAlibabaLLAMA3.1-8B-ITLlama3.1-8B-IT (Instruct)MetaQWEN3-4BQwen3-4BAlibabaLLAMA3.3-70B-ITLlama3.3-70B-IT (Instruct)MetaDEEPSEEK-R1-LLAMA-8BDeepSeek-R1-Dist-Llama-8BDeepSeekGEMMA-2-2B-ITGemma-2-2B-ITGoogle DeepmindGEMMA-2-9B-ITGemma-2-9B-ITGoogle DeepmindLLAMA3.1-8BLlama3.1-8B (Base)MetaGEMMA-2-2BGemma-2-2BGoogle DeepmindGEMMA-2-9BGemma-2-9BGoogle DeepmindP70M-DPythia-70M-DedupedEleutherAIGPT2-SMALLGPT2-SmallOpenAIJump ToJump to Source/SAEMODEL20-gemmascope-res-16kSource/SAEGoJump to FeatureMODELSource/SAEINDEXGoJump to RandomRandomGraphVisualize and trace the internal reasoning steps of a model with custom prompts, pioneered by Anthropic's circuit tracing papers.Try It: Circuit TracerYouTube: Guided DemoPost: Research LandscapeSteerModify model behavior by steering its activations using latents or custom vectors. Steering supports instruct (chat) and reasoning models, and has fully customizable temperature, strength, seed, etc.Try It: Gemma 2 - Cat SteeringSearchSearch over 50,000,000 latents/vectors, either by semantic similarity to explanation text, or by running custom text via inference through a model to find top matches. Try It: Search by ExplanationDocs: Search via InferenceSearch via InferenceMODELResidual Stream - 16kAll LayersSEARCHRun Example SearchRandom🌮 Food📰 News📖 Literary👯 Personal🧑💻 Programming🧑🔬 Technical🧑🏫 Academic💼 Business🧑⚖️ Legal🧑🏫 Educational🗼 CulturalAPI + LibrariesNeuronpedia hosts the world's first interpretability API (March 2024) - and all functionality is available by API or Python/TypeScript libraries. Most endpoints have an OpenAPI spec and interactive docs.API PlaygroundInspectGo in depth on each probe/latent/feature with top activations, top logits, activation density, and live inference testing. All dashboards have unique links, can be compiled into sharable lists, and supports IFrame embedding, as demonstrated here. Docs: ListsDocs: EmbedWho We AreNeuronpedia was created by Johnny Lin, an ex-Apple engineer who previously founded a privacy startup. Neuronpedia is supported by Decode Research, Open Philanthropy, the Long Term Future Fund, AISTOF, Anthropic, Manifund, and others.Get InvolvedCommunityGitHubContactUpskillCitation@miscneuronpedia, title = Neuronpedia: Interactive Reference and Tooling for Analyzing Neural Networks, year = 2023, note = Software available from neuronpedia.org, url = https://w.neuronpedia.org, author = Lin, Johnny