© 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 EvalsExportsAPI 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. 17908
    Prev
    Next
    INDEX
    Explanations

    names of geographical locations

    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
    314
    +0.14
    0.6%
    339
    +0.12
    0.6%
    659
    +0.11
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    659
    +0.14
    0.01
    232
    +0.12
    0.01
    588
    +0.11
    0.01
    Negative Logits
    lihood
    -1.00
    hh
    -0.69
     Koran
    -0.69
    cers
    -0.68
     Kafka
    -0.64
     Flavoring
    -0.62
    SHIP
    -0.60
     diligence
    -0.60
     discriminate
    -0.60
     deed
    -0.60
    POSITIVE LOGITS
    agher
    1.41
    illo
    1.34
    asso
    1.16
    abba
    1.15
    uda
    1.06
    arella
    1.05
    uth
    1.04
    ache
    1.02
    abis
    1.01
    anca
    1.00
    Activations Density 0.018%

    No Known Activations