© 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. 2911
    Prev
    Next
    INDEX
    Explanations

    names related to the author Mir

    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
    238
    +0.18
    0.8%
    571
    +0.12
    0.5%
    288
    +0.11
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    238
    +0.18
    0.02
    275
    +0.12
    0.01
    571
    +0.11
    0.01
    Negative Logits
    ential
    -0.63
     Primordial
    -0.62
     Combine
    -0.62
    YC
    -0.61
    CAST
    -0.61
    ļéĨĴ
    -0.60
     Merrill
    -0.60
    OPLE
    -0.60
     LCS
    -0.59
    ggies
    -0.58
    POSITIVE LOGITS
    rored
    1.31
    rors
    1.31
    acles
    1.28
    acle
    1.21
    abella
    1.11
    ror
    1.08
    abilia
    1.03
    iam
    0.96
    anda
    0.94
    abel
    0.94
    Activations Density 0.017%

    No Known Activations