© 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. 2659
    Prev
    Next
    INDEX
    Explanations

    information about metallurgy and steel processing

    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
    79
    +0.15
    0.7%
    373
    +0.10
    0.5%
    89
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    373
    +0.15
    -0.01
    232
    +0.10
    0.03
    406
    +0.09
    0.03
    Negative Logits
     canal
    -0.63
     maize
    -0.61
     bulb
    -0.58
     banana
    -0.57
     coc
    -0.56
     averaging
    -0.56
     capac
    -0.55
     envelope
    -0.54
     cloudy
    -0.54
     hindsight
    -0.54
    POSITIVE LOGITS
    addle
    0.77
    raid
    0.74
    urgy
    0.73
    roy
    0.68
    lyak
    0.66
    orage
    0.66
    ritic
    0.65
    ouver
    0.65
    mitt
    0.64
    ocratic
    0.64
    Activations Density 0.135%

    No Known Activations