© 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 EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    3. Joseph Bloom · Feature Splitting for GPT2-Small
    5. GPT2-Small
    7. Residual Feature Splitting 24576
    9. 8-RES_FS24576-JB
    11. 6345
    Prev
    Next
    INDEX
    Explanations

    keywords related to narration or storytelling, especially focusing on specific scenes or viewpoints

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_fs24576-jb
    Configuration
    jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_24576
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    24,576
    Data Type
    torch.float32
    Hook Point
    blocks.8.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    Skylion007/openwebtext
    Hook Point Layer
    8
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    373
    +0.13
    0.6%
    88
    +0.10
    0.5%
    182
    +0.10
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    88
    +0.13
    0.02
    466
    +0.10
    0.02
    446
    +0.10
    0.02
    Negative Logits
     Cipher
    -0.69
    Brave
    -0.69
    yss
    -0.67
    nces
    -0.65
    MON
    -0.64
    reddits
    -0.62
    apons
    -0.62
    adelphia
    -0.61
    é¾
    -0.61
    bleacher
    -0.60
    POSITIVE LOGITS
     view
    0.82
     differentiation
    0.79
     contention
    0.79
     reference
    0.77
     entry
    0.73
     weakness
    0.71
     vantage
    0.70
    sale
    0.69
     departure
    0.69
     contact
    0.69
    Activations Density 0.053%

    No Known Activations