© 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. DeepSeek-R1-Dist-Llama-8B
    5. 21-LLAMASCOPE-SLIMPJ-OPENR1-RES-32K
    7. 27328
    Prev
    Next
    INDEX
    Explanations

    this neuron seems to be looking for text where the author is thinking or reflecting upon something, as shown by activation on phrases like "I can," "maybe," "I think," and "was."

    oai_token-act-pair · gemini-2.0-flash
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    fnlp/Llama-Scope-R1-Distill/400M-Slimpajama-400M-OpenR1-Math-220k/L21R
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Hzfinfdu/SlimPajama-3B and open-r1/OpenR1-Math-220k
    Features
    32,768
    Data Type
    float32
    Hook Name
    blocks.21.hook_resid_post
    Architecture
    jumprelu
    Context Size
    1,024
    Dataset
    cerebras/SlimPajama-627B
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    ç¿Ķ
    -0.07
    apolis
    -0.06
    raig
    -0.06
    alance
    -0.06
     Respect
    -0.06
    elocity
    -0.06
    archy
    -0.06
    ÑĢÑı
    -0.06
    ivol
    -0.06
     Witness
    -0.06
    POSITIVE LOGITS
     recognition
    0.39
     recognize
    0.37
     Recognition
    0.36
    recogn
    0.36
     Recogn
    0.35
     recognizing
    0.34
    Recogn
    0.33
     recognized
    0.33
     recognizes
    0.33
     recogn
    0.32
    Activations Density 0.069%

    No Known Activations