© 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. 4629
    Prev
    Next
    INDEX
    Explanations

    phrases related to unspecified reasons or unknown circumstances

    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
    362
    +0.12
    0.6%
    316
    +0.11
    0.5%
    569
    +0.11
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    362
    +0.12
    0.02
    316
    +0.11
    0.02
    569
    +0.11
    0.02
    Negative Logits
    acus
    -0.86
    osate
    -0.82
    ushima
    -0.73
    nea
    -0.72
    eries
    -0.71
    byss
    -0.71
    alog
    -0.70
    Registration
    -0.70
    aneers
    -0.70
    GPU
    -0.70
    POSITIVE LOGITS
     amount
    1.35
     amounts
    1.17
     quantity
    1.14
     number
    1.03
     unspecified
    1.00
     quantities
    0.98
     timeframe
    0.91
     length
    0.88
     sums
    0.87
     undisclosed
    0.86
    Activations Density 0.062%

    No Known Activations