© 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. Google DeepMind · Exploring Gemma 2 with Gemma Scope
    5. Gemma-2-9B
    7. Residual Stream - 16k
    9. 20-GEMMASCOPE-RES-16K
    11. 6668
    Prev
    Next
    INDEX
    Explanations

    mathematical or scientific notation elements and variables

    oai_token-act-pair · gpt-4o-miniTriggered by @bot

    mathematical or computational syntax (like programming code, mathematical notation, or statistical expressions).

    oai_token-act-pair · claude-3-5-haiku-20241022Triggered by @neilrathi

    variable names, numerical identifiers, and programming syntax in code or mathematical expressions.

    oai_token-act-pair · claude-3-7-sonnet-20250219Triggered by @neilrathi

    technical mathematical/code notation with many numeric literals (often decimals) and operators (e.g., sqrt, exponent/power), including LaTeX or table/matrix syntax.

    oai_token-act-pair · gpt-5Triggered by @jyhe0408

    Looking at the activations, this neuron activates strongly on: - Mathematical expressions with numbers (particularly "0", "1", "2", "7" followed by closing delimiters) - Code-like syntax with underscores (e.g., "is_nac

    oai_token-act-pair · claude-4-5-sonnetTriggered by @jyhe0408
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GEMMA-2-9B @ 20-gemmascope-res-16k
    Configuration
    google/gemma-scope-9b-pt-res/layer_20/width_16k/average_l0_68
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    16,384
    Data Type
    float32
    Hook Name
    blocks.20.hook_resid_post
    Hook Layer
    20
    Architecture
    jumprelu
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    TagMode
    -0.62
     مشين
    -0.61
     utafitiHapana
    -0.58
     Infórmanos
    -0.57
    تقاوى
    -0.53
     Wikimedijinoj
    -0.51
    UnusedPrivate
    -0.47
    __(/*!
    -0.47
    elemField
    -0.47
    transQ
    -0.46
    POSITIVE LOGITS
     houſe
    0.39
     colazione
    0.39
     amitié
    0.38
     pleaſure
    0.36
     directos
    0.36
     pierna
    0.36
     Majefty
    0.36
     offerta
    0.36
    Toponymie
    0.35
     muligt
    0.34
    Activations Density 0.299%

    No Known Activations