© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Jacobian LensNEW
    Natural Language
    Autoencoders
    NEW
    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsinterp-engineNEWAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B
    3. 31-GEMMASCOPE-2-RES-262K
    4. 57828
    Prev
    Next
    INDEX
    Explanations

    descriptive pairs

    np_acts-logits-general · gemini-2.5-flash-lite

    aspect ratio `--ar`**Reasoning:**The `MAX_ACTIVATING_TOKENS` list contains "ar" multiple times.The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` list shows "`" (a backtick) frequently appearing after "ar".The `TOP_ACTIVATING_TEXTS` list contains numerous examples of `--ar 16:9`, `--ar 3:2`, `--ar 1:1`, and similar patterns related to "Aspect Ratio" in prompts for AI image generators like Midjourney. This strongly suggests the neuron is focused on detecting and activating for the concept of "aspect ratio" or the syntax used to specify it."Cast" is also present, and "Castling" is mentioned in the context of chess, but the "ar" pattern is far more dominant and consistently aligned with a specific technical concept demonstrated across multiple texts. The explanation must be concise and capture the primary pattern

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-pt/resid_post/layer_31_width_262k_l0_medium
    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
     Audible
    0.48
     فونٹ
    0.45
    وزی
    0.44
    ودی
    0.42
     зараз
    0.41
    ብስ
    0.41
    ကြောင်း
    0.41
     مراکز
    0.40
     solicita
    0.39
    દાવાદ
    0.39
    POSITIVE LOGITS
    your
    0.50
     your
    0.49
    B
    0.48
     you
    0.46
    你的
    0.42
    ['-
    0.42
     respectively
    0.42
     thereof
    0.42
     accordingly
    0.41
    在你
    0.41
    Activations Density 0.000%

    No Known Activations