© 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 EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Qwen3.5-4B
    3. 15-RES-MATRYOSHKA-65K
    4. 20206
    Prev
    Next
    INDEX
    Explanations

    numbers, especially when followed by specific tokens like "Jean" or "Woman", or when part of coding structures.This neuron detects numbers and specific names or titles.The core task is to find a *pattern* and express it as a short phrase.- `MAX_ACTIVATING_TOKENS`: `0, g, 5, 2, 7, 0, 5, 1, 1, 0` (mostly digits, one 'g')- `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `x, Jean, Woman, -B, ,, (` (potential words like 'Jean', 'Woman', symbols)Let's look at `TOP_ACTIVATING_TEXTS` for context when these tokens might appear:- `sizeofFPdbleword = 8; aggr FPdbleword { 'F' 0 x; { 'U' 0 lo; 'U'` : Here `0` is followed by `x`.- `,300, scoring in 1.13.1kr. Tommerre Marjea (7g Jeanbat du Vivier-Jharia du Ham) was second for J` : Here `7` is followed by `g` (which is in `MAX_ACTIVATING_TOKENS`).This neuron detects sequences of numbers, sometimes followed by names or specific identifiers

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    decoderesearch/qwen-3.5-saes/qwen-3.5-4b
    Prompts (Dashboard)
    16,384 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    åħ·æľī丰å¯ĮçļĦ
    -0.06
    计æĹ¶
    -0.06
    vog
    -0.06
    ianos
    -0.06
    æĢ§å¼º
    -0.06
    æĴħ
    -0.06
    åħ·æľīä¸Ģå®ļçļĦ
    -0.06
    asel
    -0.06
    registro
    -0.06
    ócr
    -0.06
    POSITIVE LOGITS
    å¤ļçļĦ
    0.06
    urt
    0.05
    ä¸įæĶ¹
    0.05
    ÂŃ
    0.05
    ares
    0.05
    én
    0.05
    ecome
    0.05
    æĺ¥æ±Ł
    0.05
    dict
    0.05
    -
    0.05
    Activations Density 0.024%

    No Known Activations