© 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. 4523
    Prev
    Next
    INDEX
    Explanations

    references to political candidates and their actions or positions

    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
    86
    +0.10
    0.5%
    474
    +0.10
    0.5%
    446
    +0.10
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    64
    +0.10
    -0.20
    490
    +0.10
    0.18
    86
    +0.10
    0.18
    Negative Logits
     enqu
    -0.85
    layer
    -0.77
    guide
    -0.70
    prototype
    -0.69
     decomp
    -0.69
    rower
    -0.69
    mx
    -0.68
    vertisements
    -0.67
    emouth
    -0.67
     realise
    -0.67
    POSITIVE LOGITS
     Dems
    1.06
    Republicans
    1.04
    Democrats
    0.99
     Democratic
    0.98
     Pence
    0.96
     GOP
    0.96
    Democratic
    0.93
     Mattis
    0.93
     Republican
    0.92
     Democrats
    0.91
    Activations Density 20.268%

    No Known Activations