© 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
    3. Qwen3-4B
    5. 32-TRANSCODER-HP
    7. 41995
    Prev
    Next
    INDEX
    Explanations

    code and licenses

    np_max-act-logits · gemini-2.0-flash

    Critical punctuation marks and keywords in software license boilerplates, particularly before nested clauses or critical legal terms.

    eleuther_acts_top20 · deepseek-v3Triggered by @finnwit

    comma tokens within software license/copyright headers, particularly ones appearing in standardized boilerplate sections about legal compliance or licensing terms.

    oai_token-act-pair · deepseek-v3Triggered by @finnwit
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    mwhanna/qwen3-4b-transcoders/layer_32.safetensors
    Prompts (Dashboard)
    16,384 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    163,840
    Data Type
    float32
    Hook Name
    blocks.32.mlp.hook_in
    Architecture
    transcoder
    Context Size
    8,192
    Dataset
    monology/pile-uncopyrighted
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     learnt
    -0.34
     "
    -0.31
    deal
    -0.31
     resolved
    -0.30
     spread
    -0.30
     choice
    -0.30
     permitted
    -0.29
     relev
    -0.29
    's
    -0.28
     solution
    -0.28
    POSITIVE LOGITS
     ”↵↵
    0.36
    ’↵↵
    0.33
    “My
    0.31
    è¿Ļ个人
    0.31
    “She
    0.30
    “This
    0.30
    è¿Ļæĺ¯ä¸Ģ个
    0.29
    ’a
    0.29
    )”
    0.29
    ’.
    0.29
    Activations Density 0.010%

    No Known Activations