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. 20-GEMMASCOPE-RES-16K
    6. 2296
    Prev
    Next
    INDEX
    Explanations

    statistical performance metrics in sports contexts.

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

    descriptions of sports matches or events.

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot

    phrases indicating something did not go as expected or planned, particularly in a negative way. It activates strongly for phrases like "was unable to", "were eliminated", "was forced out", "avoid being", "was bitten", "is destroyed", and "got pushed around" - all of

    oai_token-act-pair · claude-3-5-sonnet-20240620Triggered by @bot

    Following "was" or "being" and a passive verb

    np_acts-logits-general · gemini-2.0-flash

    passive negative events

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GEMMA-2-2B @ 20-gemmascope-res-16k
    Configuration
    google/gemma-scope-2b-pt-res/layer_20/width_16k/average_l0_71
    Prompts (Dashboard)
    36,864 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
     StatefulWidget
    -0.59
     reç
    -0.56
     incroy
    -0.56
     énergé
    -0.54
    <bos>
    -0.53
     hunne
    -0.53
     motivadoras
    -0.53
     bewegen
    -0.52
     internationaux
    -0.51
     modalités
    -0.50
    POSITIVE LOGITS
     disambiguazione
    0.90
     overtaken
    0.89
     attacked
    0.85
     beat
    0.84
     overrun
    0.82
     hijack
    0.78
     beaten
    0.78
     assaulted
    0.72
     trampled
    0.72
     hit
    0.71
    Activations Density 0.282%

    No Known Activations