© 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. Google DeepMind · Exploring Gemma 2 with Gemma Scope
    3. Gemma-2-9B-IT
    4. Residual Stream - 131k
    5. 20-GEMMASCOPE-RES-131K
    6. 7857
    Prev
    Next
    INDEX
    Explanations

    mathematical expressions and numerical data

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

    numbers and mathematical expressions2. statistical notation ($p<$, significances)3. code and programming constants (hexadecimal, variable declarations)The `MAX_ACTIVATING_TOKENS` list contains digits and symbols often found in these contexts. The `TOP_POSITIVE_LOGITS` reinforce the technical/programming/mathematical theme.The neuron seems to be activated by numerical data, mathematical operations, or programming syntax that involves numbers and symbols.Combining these observations, a good concise explanation would be: **numbers, math, and code**.Let's check constraints:- 3 to 20 words: "numbers, math, and code" is 4 words.- Concise explanation.- No "tokens", "patterns", "detects/predicts", "This neuron".- Not capitalized unless proper noun.- Specific.This seems to fit. numbers, math, and code

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GEMMA-2-9B-IT @ 20-gemmascope-res-131k
    Configuration
    google/gemma-scope-9b-it-res/layer_20/width_131k/average_l0_81
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    131,072
    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
    transQ
    -0.70
    Rüyada
    -0.56
    SourceChecksum
    -0.55
     kaarangay
    -0.53
    verwijspagina
    -0.48
     typelib
    -0.48
     propOrder
    -0.48
     שוליים
    -0.47
    endphp
    -0.45
    tagext
    -0.45
    POSITIVE LOGITS
     ivelany
    0.51
    HasAnnotation
    0.40
    RTGC
    0.38
    WriteTagHelper
    0.37
     constituents
    0.37
     harmonic
    0.37
    STUD
    0.36
    waitKey
    0.35
    addContainerGap
    0.34
     LIVES
    0.34
    Activations Density 0.014%

    No Known Activations