© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports Community BlogPrivacy & TermsContact
    1. Home
    2. Google DeepMind · Exploring Gemma 2 with Gemma Scope
    3. Gemma-2-9B-IT
    4. Residual Stream - 131k
    5. 20-GEMMASCOPE-RES-131K
    6. 32098
    Prev
    Next
    INDEX
    Explanations

    sections of text that are excessively repetitive or lack new content

    oai_token-act-pair · gpt-4o-miniTriggered by @bot

    useruseruseruseruseruser</TOKENS_AFTER_MAX_ACTIVATING_TOKEN><TOP_POSITIVE_LOGITS>########.enfanstagHelperRunnerrecevrezcirconstancesâmesméribaixEuropy</TOP_POSITIVE_LOGITS><TOP_ACTIVATING_TEXTS> Community center 6 -0.03 -0.48; 0.41       Other 5 -0.24 -0. guidance receptor 2 0.2710 TNC tenascin C Positive regulation of apoptotic process 1.30E-04 0.410<end_of_turn> 5 or more 7 0.84 -0.48; 0.65   Educators           11.0 18 13 18 3.8 1919+21 1.337 15 89 19-macroglobulin 0.2780 CSTA cystatin A 0.1772 CST6 cystatin E/M<end_of_turn> 81 (18.7) 107 (31.1) \- \-  • 5 4 (0.9) 17 (4.9) <start_of_turn>user @rheol8; @rheol9; @rheol10; @rheol11<start_of_turn>user => { expect( astNode.nodeToValue(methodDefinition2) ).toBe('<anonymous<start_of_turn>user [^2]: The author, Ethel Gilbert, is the mother of Peter Gilbert who is a contributor to<start_of_turn>user the dynamic change at all between Dinah and her boss? "They've always been close so there'<start_of_turn>user const variableDeclarator2 = parse('var foo;').declarations[0]; it('</TOP_ACTIVATING_TEXTS>Explain the neuron above with a word or phrase, not a complete sentence. community, educators, guidance

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GEMMA-2-9B-IT @ 20-gemmascope-res-131k
    Configuration
    google/gemma-scope-9b-it-res/layer_20/width_131k/average_l0_81
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    131,072
    Data Type
    float32
    Hook Name
    blocks.20.hook_resid_post
    Hook Layer
    20
    Architecture
    jumprelu
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    Personendaten
    -0.88
    webElementXpaths
    -0.70
    MLLoader
    -0.68
    UnusedPrivate
    -0.66
    GEBURTSDATUM
    -0.66
    transQ
    -0.65
    Hochspringen
    -0.63
    sidemargin
    -0.63
    󠁢
    -0.62
     snippetHide
    -0.61
    POSITIVE LOGITS
    ########.
    0.55
     enfans
    0.51
    tagHelperRunner
    0.45
     recevrez
    0.41
     circonstances
    0.37
     âmes
    0.37
     méri
    0.36
    ↵↵
    0.35
     baix
    0.35
     Europy
    0.34
    Activations Density 0.000%

    No Known Activations