© 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. 3444
    Prev
    Next
    INDEX
    Explanations

    instances of complex character dynamics and relationships

    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
    104
    +0.11
    0.5%
    644
    +0.10
    0.5%
    660
    +0.08
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    421
    +0.11
    0.07
    673
    +0.10
    0.08
    346
    +0.08
    0.08
    Negative Logits
     respectively
    -0.64
    bery
    -0.64
    geries
    -0.63
     allocations
    -0.63
     hybrids
    -0.62
     supermarkets
    -0.62
     interchangeable
    -0.62
    Engineers
    -0.62
     competition
    -0.62
     classification
    -0.60
    POSITIVE LOGITS
    itud
    0.73
    jad
    0.72
    aste
    0.70
    vae
    0.70
    cook
    0.70
    agna
    0.70
    raged
    0.69
    ctica
    0.67
     herself
    0.66
     Unsure
    0.66
    Activations Density 1.938%

    No Known Activations