© 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. Under Peer Review · Sparse Autoencoders for Pythia-70M-Deduped
    5. Pythia-70M-Deduped
    7. Residuals
    9. 2-RES-SM
    11. 27867
    Prev
    Next
    INDEX
    Explanations

    instances of the word "attention" in various contexts

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    ctigges/pythia-70m-deduped__res-sm_processed/2-res-sm
    Prompts (Dashboard)
    32,768 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    32,768
    Data Type
    torch.float32
    Hook Name
    blocks.2.hook_resid_post
    Hook Layer
    2
    Architecture
    standard
    Context Size
    128
    Dataset
    EleutherAI/the_pile_deduplicated
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    287
    +0.15
    0.8%
    38
    +0.15
    0.8%
    466
    +0.13
    0.7%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    287
    +0.15
    0.04
    38
    +0.15
    0.02
    306
    +0.13
    0.02
    Negative Logits
    Ľ
    -3.38
    ĻĤ
    -2.89
    Īĺ
    -2.88
    ı
    -2.71
    Ħ
    -2.71
    Ģ
    -2.65
    ª
    -2.62
    IJ
    -2.57
    ķ
    -2.56
    ³
    -2.55
    POSITIVE LOGITS
     paid
    1.75
    al
    1.71
    piece
    1.65
    pieces
    1.63
    worthy
    1.61
     accorded
    1.60
     fatigue
    1.58
    worth
    1.57
    acted
    1.56
    attention
    1.56
    Activations Density 0.188%

    No Known Activations