Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsExportsSlackBlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Google DeepMind · Exploring Gemma 2 with Gemma Scope
    3. Gemma-2-2B
    4. Residual Stream - 16k
    5. 25-GEMMASCOPE-RES-16K
    6. 10391
    Prev
    Next
    INDEX
    Explanations

    phrases related to high-profile lifestyle and luxury

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

    say "▁" words

    np_max-act-logits · gpt-4o-miniTriggered by @arientigiacomo2003

    proper nouns and named entities, especially locations, organizations, and people.

    oai_token-act-pair · gpt-4.1-mini-2025-04-14Triggered by @arientigiacomo2003

    The highlighted sequences predominantly consist of proper nouns, place names, organizations, and specific entities—often including geographic locations, institutional names, and branded terms. They frequently appear in contexts indicating attribution, location, or formal titles. The patterns show a tendency to mark multi-word and compound named entities, especially those combining place identifiers with organizations or events.

    eleuther_acts_top20 · gpt-4.1-mini-2025-04-14Triggered by @arientigiacomo2003

    US/American

    np_max-act-logits · gpt-4.1-mini-2025-04-14Triggered by @arientigiacomo2003

    entities of societal importance, such as notable people, places, organizations, and major topics affecting society.

    oai_token-act-pair · deepseek-r1-dist-llama-70bTriggered by @arientigiacomo2003
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GEMMA-2-2B @ 25-gemmascope-res-16k
    Configuration
    google/gemma-scope-2b-pt-res/layer_25/width_16k/average_l0_116
    Prompts (Dashboard)
    36,864 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    16,384
    Data Type
    float32
    Hook Name
    blocks.25.hook_resid_post
    Hook Layer
    25
    Architecture
    jumprelu
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    (!__
    -0.66
     cherchés
    -0.63
     ***!
    -0.55
    ttemberg
    -0.55
     bezeichneter
    -0.54
    arakhand
    -0.52
    útbol
    -0.51
    CPtr
    -0.50
     pylori
    -0.50
    MigrationBuilder
    -0.49
    POSITIVE LOGITS
     US
    2.06
     American
    1.94
     Americans
    1.82
    American
    1.76
     USA
    1.74
     american
    1.72
     amerik
    1.71
     America
    1.71
    US
    1.69
     amerikan
    1.63
    Activations Density 0.401%

    No Known Activations