© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Natural Language
    Autoencoders
    NEW    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    3. OpenAI · Sparse Autoencoder for GPT2-Small - v5 2024
    5. GPT2-Small
    7. Attention Out - 32k
    9. 3-ATT_32K-OAI
    11. 3082
    Prev
    Next
    INDEX
    Explanations

    references to grabbing or holding objects

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 3-att_32k-oai
    Configuration
    jbloom/GPT2-Small-OAI-v5-32k-attn-out-SAEs/v5_32k_layer_3
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    32,768
    Data Type
    torch.float32
    Hook Name
    blocks.3.hook_attn_out
    Hook Layer
    3
    Architecture
    standard
    Context Size
    64
    Dataset
    Skylion007/openwebtext
    Activation Function
    topk
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    554
    +0.12
    0.5%
    741
    +0.10
    0.5%
    304
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    587
    +0.12
    0.02
    437
    +0.10
    0.02
    533
    +0.09
    0.02
    Negative Logits
    appa
    -0.78
    rome
    -0.72
    cair
    -0.71
    erent
    -0.70
    ople
    -0.70
    amiya
    -0.70
    phans
    -0.69
    raud
    -0.69
    gallery
    -0.68
    ropolis
    -0.67
    POSITIVE LOGITS
     hats
    0.71
     concess
    0.68
     vacated
    0.68
    ÙĴ
    0.67
    hold
    0.66
     attention
    0.65
     coveted
    0.64
     foothold
    0.62
     Sacred
    0.60
     GEAR
    0.60
    Activations Density 0.135%

    No Known Activations