© 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. 13906
    Prev
    Next
    INDEX
    Explanations

    personal impressions and feelings expressed by individuals

    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
    702
    +0.10
    0.5%
    274
    +0.10
    0.4%
    531
    +0.10
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    274
    +0.10
    0.04
    737
    +0.10
    0.04
    355
    +0.10
    0.03
    Negative Logits
    aples
    -0.68
     belonging
    -0.67
    orks
    -0.65
     oneself
    -0.64
     yourselves
    -0.62
    necessary
    -0.61
    pire
    -0.60
    rir
    -0.60
    thinkable
    -0.60
     Klux
    -0.58
    POSITIVE LOGITS
     boils
    0.79
    nings
    0.75
     recommendation
    0.75
    çīĪ
    0.74
    favorite
    0.71
     takeaway
    0.71
     foray
    0.69
     consisted
    0.68
    elist
    0.68
     sucks
    0.68
    Activations Density 0.164%

    No Known Activations