© 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. 4318
    Prev
    Next
    INDEX
    Explanations

    technical terms and goals related to software development and project implementation

    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
    381
    +0.11
    0.5%
    745
    +0.10
    0.5%
    737
    +0.10
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    170
    +0.11
    0.03
    650
    +0.10
    0.03
    580
    +0.10
    0.03
    Negative Logits
     notes
    -0.65
    ython
    -0.63
    eor
    -0.62
     slips
    -0.62
     equivalents
    -0.62
     clips
    -0.62
    shit
    -0.62
     hints
    -0.61
     notices
    -0.61
     stains
    -0.60
    POSITIVE LOGITS
     maximizing
    0.99
     restoring
    0.82
     ensuring
    0.79
     revenge
    0.78
     fulfilled
    0.77
     ambitious
    0.77
     fostering
    0.76
     preserving
    0.75
     maximize
    0.75
     vengeance
    0.74
    Activations Density 0.167%

    No Known Activations