© 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. 9-GEMMASCOPE-RES-131K
    6. 78512
    Prev
    Next
    INDEX
    Explanations

    chemical elements and compounds, particularly those related to iron and its oxidation states

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

    iron(II The `MAX_ACTIVATING_TOKENS` list contains only '2'.The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` list contains '~' and '+'.The `TOP_ACTIVATING_TEXTS` list shows many examples of `Fe^2+^`, `Fe^3+^`, and `Fe~2~`.The `TOP_POSITIVE_LOGITS` list contains technical terms and some Arabic words.The overwhelming pattern is the presence of chemical formulas, specifically iron(II) and iron(III) ions, and sometimes other ions like Europium(II).The token '2' in `MAX_ACTIVATING_TOKENS` likely corresponds to the '2' in `Fe^2+^` or `Fe~2~`.The '~' and '+' in `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` likely correspond to the subscript notation or the plus charge in the ions.Therefore, the neuron seems to detect or predict patterns related to such chemical species.Explanation: iron(II) and iron(III) ions

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GEMMA-2-9B-IT @ 9-gemmascope-res-131k
    Configuration
    google/gemma-scope-9b-it-res/layer_9/width_131k/average_l0_121
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    131,072
    Data Type
    float32
    Hook Name
    blocks.9.hook_resid_post
    Hook Layer
    9
    Architecture
    jumprelu
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     kokona
    -0.33
     juara
    -0.31
     lengkap
    -0.29
     tarjo
    -0.29
     avulla
    -0.28
     ganze
    -0.28
     brought
    -0.28
     työ
    -0.27
     complètes
    -0.27
     henkil
    -0.26
    POSITIVE LOGITS
     NSCoder
    0.75
    AddTagHelper
    0.71
     تضيفلها
    0.64
    FailureListener
    0.64
     rhestr
    0.63
    RefNanny
    0.61
    tagHelperRunner
    0.60
    MessageTagHelper
    0.59
    complexContent
    0.58
     للمعارف
    0.57
    Activations Density 0.221%

    No Known Activations