© 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. 10975
    Prev
    Next
    INDEX
    Explanations

    dates in August

    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
    7
    +0.13
    0.6%
    746
    +0.11
    0.5%
    549
    +0.11
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    701
    +0.13
    0.01
    7
    +0.11
    0.01
    254
    +0.11
    0.01
    Negative Logits
     Messi
    -0.74
     TYPE
    -0.72
     Nadu
    -0.70
    BO
    -0.69
    TPS
    -0.65
     Oo
    -0.63
     Leopard
    -0.63
     Ronaldo
    -0.62
    Redditor
    -0.62
    zzi
    -0.61
    POSITIVE LOGITS
    uration
    1.47
    urated
    1.40
    urations
    1.40
    urate
    1.29
    ural
    1.22
    mentation
    1.10
    ust
    1.07
    sburg
    1.04
    roup
    1.01
    ured
    0.99
    Activations Density 0.020%

    No Known Activations