© 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 EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-12B-IT
    3. 12-GEMMASCOPE-2-RES-16K
    4. 14182
    Prev
    Next
    INDEX
    Explanations

    The dominant token is 'が' (Japanese particle) and its Korean equivalent '가'.Let's look for patterns related to these particles.**TOP_POSITIVE_LOGITS:**- ために (tame ni - in order to, for the sake of)- 아니라 (aranira - not only)- 아닌 (anin - not)- }$ (symbol)- ️⃣ (symbol)- 4 (number)- 7 (number)- ১৮ (number, Bengali)- CTURE (part of a word)- ’ (apostrophe)This list seems very mixed, containing Japanese phrases, symbols, numbers, and even a Bengali digit. It doesn't immediately highlight a strong semantic pattern related to 'が'/'가'.**TOP_ACTIVATING_TEXTS:**- `. * \`o\` (を): Marks the direct object. * `ga` (が): often used by children or in a playful tone.This neuron detects Japanese particle `ga` followed by grammatical elements or sentence endings

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-12b-it/resid_post/layer_12_width_16k_l0_medium
    Prompts (Dashboard)
    238,145 prompts, 512 tokens each
    Dataset (Dashboard)
    lmsys + oasst1
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    o
    3.30
    ו
    2.86
    р
    2.78
    л
    2.78
    ро
    2.34
    a
    2.27
    ある
    2.13
    ो
    2.05
    ва
    1.97
    र
    1.95
    POSITIVE LOGITS
    ために
    2.03
     아니라
    2.03
     아닌
    1.96
    }$
    1.84
    ️⃣
    1.79
    4
    1.76
    7
    1.75
    ১৮
    1.74
    CTURE
    1.72
    ’
    1.67
    Activations Density 0.004%

    No Known Activations