© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports Community BlogPrivacy & TermsContact
    1. Home
    3. Joseph Bloom · Feature Splitting for GPT2-Small
    5. GPT2-Small
    7. Residual Feature Splitting 12288
    9. 8-RES_FS12288-JB
    11. 9716
    Prev
    Next
    INDEX
    Explanations

    the word "replace" or related words indicating substitution

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_fs12288-jb
    Configuration
    jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_12288
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    12,288
    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
    507
    +0.11
    0.5%
    143
    +0.10
    0.4%
    166
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    507
    +0.11
    0.03
    123
    +0.10
    0.02
    735
    +0.09
    0.02
    Negative Logits
    raq
    -0.95
    insula
    -0.78
    hest
    -0.73
     Reasons
    -0.68
    jah
    -0.65
    eca
    -0.63
    reprene
    -0.62
    about
    -0.61
     isEnabled
    -0.61
    estate
    -0.60
    POSITIVE LOGITS
    atives
    0.76
     obsolete
    0.74
    utions
    0.73
     retiring
    0.73
     redundant
    0.72
    ments
    0.70
    itute
    0.70
     bulbs
    0.70
    mented
    0.68
     scratch
    0.68
    Activations Density 0.106%

    No Known Activations