© 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
    3. Gemma-2-2B
    5. 22-GEMMASCOPE-TRANSCODER-16K
    7. 10875
    Prev
    Next
    INDEX
    Explanations

    This neuron seems to have picked up a variety of patterns: Japanese sentence endings, the phrase "Do Cmd", adjectives, and something about "Long". I'm going to guess it's looking for something that might be used in source code or programming documentation. It finds elements of code snippets

    oai_token-act-pair · gemini-2.0-flash

    Technical writing

    np_max-act-logits · gemini-2.0-flash
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2b-pt-transcoders/layer_22/width_16k/average_l0_15
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    16,384
    Data Type
    float32
    Hook Name
    blocks.22.ln2.hook_normalized
    Architecture
    jumprelu_transcoder
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    .
    -0.60
     
    -0.57
     you
    -0.56
    ,
    -0.52
    aarrggbb
    -0.49
     You
    -0.47
     Am
    -0.45
    ResumeLayout
    -0.45
     aj
    -0.44
    IDES
    -0.44
    POSITIVE LOGITS
     ་་
    0.89
     سكانية
    0.81
     iſt
    0.76
     purpoſe
    0.75
     XNUMX
    0.74
     ―――――
    0.74
     dieß
    0.71
     gddr
    0.71
    ՚
    0.70
     Whig
    0.70
    Activations Density 0.512%

    No Known Activations