© 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. 4-RES-SM
    11. 14658
    Prev
    Next
    INDEX
    Explanations

    punctuation marks and their placement in the text

    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/4-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.4.hook_resid_post
    Hook Layer
    4
    Architecture
    standard
    Context Size
    128
    Dataset
    EleutherAI/the_pile_deduplicated
    Activation Function
    relu
    Embeds
    IFrame
    Link
    •probe features
    periods, esp. in commerical text

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    307
    +0.29
    1.8%
    348
    +0.16
    1.0%
    306
    +0.14
    0.9%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    400
    +0.29
    0.06
    258
    +0.16
    0.11
    307
    +0.14
    0.11
    Negative Logits
    "}](#
    -1.90
     wrong
    -1.81
     insult
    -1.59
     upside
    -1.54
    ietz
    -1.49
     passages
    -1.47
    ercase
    -1.46
    cliffe
    -1.45
     culprit
    -1.42
    apine
    -1.42
    POSITIVE LOGITS
    ĥ½
    2.92
    2.36
      ↵
    2.36
    <|outofrange|>
    2.36
    ↵↵
    2.36
     ↵
    2.36
     ↵
    2.36
    ↵
    2.36
    ↵
    2.36
             ↵
    2.36
    Activations Density 0.779%

    No Known Activations