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

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsExportsSlackBlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Gemma-3-27B
    3. 16-GEMMASCOPE-2-RES-16K
    4. 1
    Prev
    Next
    INDEX
    Explanations

    identifying and quantifying the overlap between upper and lower layer patterns.

    oai_token-act-pair · gpt-5-nanoTriggered by @aardvarkkr

    common words or tokens

    np_max-act · gemini-2.5-flashTriggered by @aardvarkkr

    The neuron primarily activates on frequently occurring words like "the" and "and" when they appear in technical or instructional contexts, often in close proximity to numbers or specialized terms.

    oai_token-act-pair · gemini-2.5-flashTriggered by @aardvarkkr
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-pt/resid_post
    Prompts (Dashboard)
    392,802 prompts, 256 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     
    0.57
    ,
    0.49
    ;
    0.43
     is
    0.42
    .
    0.42
     an
    0.41
    ę
    0.39
    ،
    0.39
     in
    0.37
    :
    0.37
    POSITIVE LOGITS
     데이터를
    0.41
     გამოყენ
    0.39
    υ
    0.39
     அறிவியல்
    0.38
     வகுப்பு
    0.38
     εφαρμο
    0.38
    데이터
    0.38
     يستخدم
    0.37
     χρησιμοποι
    0.37
     pemasaran
    0.37
    Activations Density 8.224%

    No Known Activations