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

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports Community BlogPrivacy & 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. 19-GEMMASCOPE-RES-16K
    6. 3297
    Prev
    Next
    INDEX
    Explanations

    phrases related to harm and the intention behind actions that cause it

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GEMMA-2-2B @ 19-gemmascope-res-16k
    Configuration
    google/gemma-scope-2b-pt-res/layer_19/width_16k/average_l0_73
    Prompts (Dashboard)
    36,864 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    16,384
    Data Type
    float32
    Hook Name
    blocks.19.hook_resid_post
    Hook Layer
    19
    Architecture
    jumprelu
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     Мексичка
    -0.64
    ագրություններ
    -0.62
     lenker
    -0.59
    Revenir
    -0.56
    сылкі
    -0.53
    tagHelper
    -0.53
     sweeter
    -0.50
    ofire
    -0.50
     Tack
    -0.49
    eschön
    -0.49
    POSITIVE LOGITS
     harm
    1.83
     Harm
    1.43
    harm
    1.37
    Harm
    1.24
     harms
    1.21
     damage
    1.21
     daño
    1.17
     hurt
    1.14
     harmed
    1.11
    damage
    1.05
    Activations Density 0.235%

    No Known Activations