© 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. Joseph Bloom · Feature Splitting for GPT2-Small
    5. GPT2-Small
    7. Residual Feature Splitting 24576
    9. 8-RES_FS24576-JB
    11. 5242
    Prev
    Next
    INDEX
    Explanations

    information related to the terms of use and copyright for an online site

    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
    563
    +0.11
    0.5%
    443
    +0.10
    0.4%
    373
    +0.10
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    567
    +0.11
    0.02
    211
    +0.10
    0.01
    39
    +0.10
    0.03
    Negative Logits
     flank
    -0.70
    ãĤ©
    -0.69
     reperto
    -0.65
     Median
    -0.65
     barracks
    -0.65
    ploy
    -0.64
     unemployment
    -0.63
    gap
    -0.62
     midterm
    -0.61
     senators
    -0.61
    POSITIVE LOGITS
     trademarks
    1.09
     disclaim
    0.98
     hereby
    0.94
    yrights
    0.92
     Attribution
    0.92
    soever
    0.90
     WARRANT
    0.90
     Sponsor
    0.86
     copyright
    0.86
     indemn
    0.85
    Activations Density 0.377%

    No Known Activations