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

    Neuronpedia

    Jacobian LensNEW
    Natural Language
    Autoencoders
    NEW    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsinterp-engineNEWAPI Community BlogPrivacy & TermsContact
    1. Home
    3. Gemma-2-27B
    5. 10-GEMMASCOPE-RES-131K
    7. 102531
    Prev
    Next
    INDEX
    Explanations

    prefix patterns ending words

    np_acts-logits-general · gemini-2.5-flash-lite

    The neuron selectively fires on sentence‐ or section‐initial content words—i.e. the first (usually capitalized) word at the start of a segment.

    oai_token-act-pair · o4-miniTriggered by @jyhe0408
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-27b-pt-res/layer_10/width_131k
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    ”
    -1.96
    <bos>
    -1.85
    6
    -1.83
    」
    -1.80
    3
    -1.77
    ".
    -1.74
    8
    -1.73
    。
    -1.69
    matical
    -1.66
    5
    -1.63
    POSITIVE LOGITS
     them
    2.38
     calitate
    1.73
    がありますが
    1.66
     vertrouwen
    1.63
     it
    1.59
     však
    1.54
    を作って
    1.53
     ought
    1.50
     tilby
    1.49
     promesse
    1.49
    Activations Density 1.650%

    No Known Activations