© 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-3-12B
    5. 24-GEMMASCOPE-2-RES-16K
    7. 3056
    Prev
    Next
    INDEX
    Explanations

    code keywords and structures

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

    This neuron fires on programming‐style tokens—especially code keywords (like “if,” “new,” “destroy,” “acts_as_list”) and associated punctuation (brackets, parentheses, colons)—marking sections of source code rather than prose.

    oai_token-act-pair · o4-miniTriggered by @jyhe0408
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-12b-pt/resid_post/layer_24_width_16k_l0_medium
    Prompts (Dashboard)
    392,802 prompts, 256 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    UV
    0.59
     humo
    0.59
    ">&
    0.58
    );
    0.57
    LIGHT
    0.57
     NodeJS
    0.56
     );
    0.55
    valueOf
    0.55
     UV
    0.53
    ):
    0.53
    POSITIVE LOGITS
    ancellor
    0.64
     elsif
    0.58
    जनिक
    0.52
     bargains
    0.51
    downcase
    0.50
    我們先
    0.49
    allenges
    0.49
     chancellor
    0.48
    anjian
    0.48
    ̀n
    0.48
    Activations Density 0.003%

    No Known Activations