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

    Neuronpedia

    Natural Language
    Autoencoders
    NEW
    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-12B
    3. 24-GEMMASCOPE-2-RES-16K
    4. 12154
    Prev
    Next
    INDEX
    Explanations

    process, config, or rec

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

    The neuron fires strongly on subword pieces of long compounds (e.g. the German “Installations-date-inamen” split into Installations, date, in, amen) or similarly on the “Ket” piece of “Ketoslim,” meaning it’s detecting subword fragments that form part of lengthy concatenated words.

    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
    y
    0.96
    sheet
    0.88
    t
    0.87
    take
    0.83
    tion
    0.82
    carry
    0.82
    tap
    0.80
    grant
    0.79
    á
    0.78
    table
    0.78
    POSITIVE LOGITS
     esforços
    0.86
     russo
    0.77
     Futebol
    0.77
    الة
    0.76
    ચાર
    0.75
     populaire
    0.75
     schermo
    0.74
     ataques
    0.72
     caution
    0.71
     cautionary
    0.71
    Activations Density 0.000%

    No Known Activations