© 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 EvalsExportsinterp-engineNEWAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Google DeepMind · Exploring Gemma 2 with Gemma Scope
    3. Gemma-2-9B
    4. Residual Stream - 131k
    5. 20-GEMMASCOPE-RES-131K
    6. 102598
    Prev
    Next
    INDEX
    Explanations

    references to planning and hosting social events or parties

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

    The marked tokens across these examples represent highly varied content types—narrative text, code, academic prose, and tabular data—with no consistent pattern. The delimiters appear to highlight contextually important elements that vary by document type: specific nouns and phrases in narratives, technical terms in scientific writing, and code constructs in programming text. Without a coherent thematic or linguistic pattern connecting the marked items across diverse genres, no meaningful generalization can be drawn from these examples.

    eleuther_acts_top20 · claude-4-5-haikuTriggered by @wallbreaker5th
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-9b-pt-res/layer_20/width_131k/average_l0_114
    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
     purged
    -0.64
    ########.
    -0.62
    BufferException
    -0.59
    UnitTesting
    -0.56
    #+#
    -0.55
    RegressionTest
    -0.54
    URLException
    -0.54
    __":
    
    -0.53
    يكب
    -0.53
     bParam
    -0.53
    POSITIVE LOGITS
     tranſ
    0.42
     child
    0.39
     papá
    0.38
     muſt
    0.37
     houſe
    0.35
     ſur
    0.34
     Photocase
    0.34
     garçons
    0.33
    yyj
    0.33
     miſ
    0.33
    Activations Density 0.030%

    No Known Activations