© 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. 20-TRANSCODER-HP
    7. 141647
    Prev
    Next
    INDEX
    Explanations

    Random, Lower-Quality text

    np_max-act · gemini-2.0-flash

    . (first token from MAX_ACTIVATING_TOKENS)

    np_max-act · deepseek-v3Triggered by @fwitzenh

    Common token sequences appearing after questions and answers, typically relate to syntax or instructions and occasionally cite references.

    eleuther_acts_top20 · deepseek-v3Triggered by @fwitzenh
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    mwhanna/qwen3-4b-transcoders/layer_20.safetensors
    Prompts (Dashboard)
    16,384 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    163,840
    Data Type
    float32
    Hook Name
    blocks.20.mlp.hook_in
    Architecture
    transcoder
    Context Size
    8,192
    Dataset
    monology/pile-uncopyrighted
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     никак
    -0.32
    没æ³ķ
    -0.29
    çŃ·
    -0.28
     nowhere
    -0.27
     unidentified
    -0.26
     lib
    -0.26
    ç®Ģä»ĭ
    -0.26
     LIB
    -0.26
     ì¡°
    -0.25
    ì¡°
    -0.25
    POSITIVE LOGITS
    edException
    0.32
    manuel
    0.27
     perfect
    0.27
    ivity
    0.26
     doorstep
    0.25
    åĪĩå°Ķ
    0.25
    annis
    0.25
    oub
    0.25
    æĭ¦
    0.25
    perfect
    0.24
    Activations Density 69.038%

    No Known Activations