© 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. Qwen3.5-9B Base
    3. 15-QWENSCOPE-RES-64K
    4. 58642
    Prev
    Next
    INDEX
    Explanations

    numbers KT279554 to KT279561"- "500 to 700 MB"- "14 days to 3 years"- "1.10(-10) to 1.10(-7]"- "14 days to 41 months"- "tens-of-thousands to tens-of-millions"- "3-week to 6-week intervals"- "T1 to T3 breast cancer"- "2006 to 2019 ADLINK"- "semi-arid to arid climate"- "sheer to intense color"The common pattern across `TOP_ACTIVATING_TEXTS` is the word "to" used to indicate a range, sequence, or limit. The `MAX_ACTIVATING_TOKENS` reinforce this with "to". While the `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` and `TOP_POSITIVE_LOGITS` are diverse, they don't contradict the range idea, and some might be related (e.g., "tens" -> "tens-of-thousands", "ranges").Therefore, the neuron seems to be related to ranges.range indicator

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    ocene
    -0.08
    ahan
    -0.08
    preso
    -0.08
    revet
    -0.08
    ¸
    -0.08
    aret
    -0.07
    èº
    -0.07
    ukka
    -0.07
    енок
    -0.07
    åĴĮåIJĮäºĭ
    -0.07
    POSITIVE LOGITS
    /of
    0.07
     seterusnya
    0.07
     غاÙĬØ©
    0.06
    Ùĭ
    0.06
    hra
    0.06
     ranges
    0.06
    /o
    0.06
    æĩ¿
    0.06
    çļĦèĮĥåĽ´
    0.06
    akedown
    0.06
    Activations Density 0.144%

    No Known Activations