© 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. MLP Post
    9. 0-MLP-SM
    11. 25855
    Prev
    Next
    INDEX
    Explanations

    instances of the word "fold" and its variants, indicating a focus on folding or bending concepts

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    ctigges/pythia-70m-deduped__mlp-sm_processed/0-mlp-sm
    Prompts (Dashboard)
    32,768 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    32,768
    Data Type
    torch.float32
    Hook Name
    blocks.0.hook_mlp_out
    Hook Layer
    0
    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
    172
    +0.14
    0.8%
    280
    +0.13
    0.7%
    347
    +0.12
    0.7%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    347
    +0.14
    0.01
    280
    +0.13
    0.01
    172
    +0.12
    0.01
    Negative Logits
    unnumbered
    -1.71
    hips
    -1.52
     Photograph
    -1.47
     Well
    -1.39
    hip
    -1.34
    jee
    -1.33
     Gardens
    -1.32
    >();
    -1.32
     mere
    -1.32
     Kit
    -1.29
    POSITIVE LOGITS
    maker
    1.80
    ings
    1.68
    idable
    1.68
    makers
    1.66
    rant
    1.57
    ahead
    1.57
    sheet
    1.53
    gens
    1.53
    able
    1.51
    making
    1.49
    Activations Density 0.008%

    No Known Activations