© 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. Qwen3-4B
    3. 25-TRANSCODER-HP
    4. 108254
    Prev
    Next
    INDEX
    Explanations

    numbers

    np_max-act-logits · gemini-2.0-flash

    numbers and mathematical sequences.

    oai_token-act-pair · deepseek-v3Triggered by @fwitzenh

    The tokens consist of word segments or prefixes that indicate numerical elements (e.g., years, dates, license versions), negations, or specific grammatical markers like comparative adjectives. These patterns relate to specific contexts such as copyright years, historical dates, mathematical sequences, or comparative phrases.

    eleuther_acts_top20 · deepseek-v3Triggered by @fwitzenh

    numerical tokens and quantities in various contexts

    oai_token-act-pair · deepseek-r1Triggered by @fwitzenh

    Explanation could not be parsed.

    eleuther_acts_top20 · deepseek-r1Triggered by @fwitzenh
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    mwhanna/qwen3-4b-transcoders/layer_25.safetensors
    Prompts (Dashboard)
    16,384 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    163,840
    Data Type
    float32
    Hook Name
    blocks.25.mlp.hook_in
    Architecture
    transcoder
    Context Size
    8,192
    Dataset
    monology/pile-uncopyrighted
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    çĿ¹
    -0.28
    æĶ¸
    -0.27
    meet
    -0.26
    -members
    -0.25
    FN
    -0.25
    åºĶæĶ¶
    -0.24
    uft
    -0.23
    PRIMARY
    -0.23
     secretly
    -0.23
    ike
    -0.23
    POSITIVE LOGITS
     less
    0.36
    ä¸įä¸Ĭ
    0.36
    ä¸įåĩº
    0.34
    iless
    0.33
     fewer
    0.33
     Less
    0.33
     neither
    0.32
     non
    0.31
    -less
    0.31
     none
    0.31
    Activations Density 31.264%

    No Known Activations