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

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExportsSlackBlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Google DeepMind · Exploring Gemma 2 with Gemma Scope
    3. Gemma-2-9B-IT
    4. Residual Stream - 131k
    5. 20-GEMMASCOPE-RES-131K
    6. 64880
    Prev
    Next
    INDEX
    Explanations

    corporate or financial terms related to agreements and partnerships

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

    established in**Reasoning:**The `MAX_ACTIVATING_TOKENS` list contains "established". The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` list contains "in" and ",". In the `TOP_ACTIVATING_TEXTS`, we see "companies established in 1881". This strongly suggests the neuron is associated with the phrase "established in".Other tokens like "start", "vein", "partnership" also appear, and their following tokens like "ted", "in", "payments" could form phrases, but "established in" is the most direct and clear pattern observed across the lists, particularly between `MAX_ACTIVATING_TOKENS` and `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`, reinforced by an example in `TOP_ACTIVATING_TEXTS`.The phrase "established in" fits the length constraints (3-20 words) and is specific. established in

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GEMMA-2-9B-IT @ 20-gemmascope-res-131k
    Configuration
    google/gemma-scope-9b-it-res/layer_20/width_131k/average_l0_81
    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
    eleste
    -0.56
     cherchés
    -0.46
     Opis
    -0.44
    Dort
    -0.43
    Opis
    -0.41
     Rumuni
    -0.41
     opis
    -0.41
     Dort
    -0.40
    Voltaje
    -0.40
    Gost
    -0.40
    POSITIVE LOGITS
    oredCriteria
    0.43
    RegistryLite
    0.41
    期刊论文
    0.39
    tagHelperRunner
    0.39
    WithIdentifier
    0.38
     viime
    0.36
    UnusedPrivate
    0.36
    PerformLayout
    0.35
    存于互联网档案馆
    0.35
    PreInfinity
    0.34
    Activations Density 0.114%

    No Known Activations