© 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. 5455
    Prev
    Next
    INDEX
    Explanations

    references to legal cases or issues

    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
    481
    +0.10
    0.5%
    100
    +0.10
    0.4%
    40
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    756
    +0.10
    -0.02
    86
    +0.10
    0.04
    150
    +0.09
    0.03
    Negative Logits
    iversary
    -0.92
    iku
    -0.91
    tainment
    -0.77
     apocalypse
    -0.76
    ãĤ·ãĥ£
    -0.73
     someday
    -0.73
     rejoice
    -0.73
     tomorrow
    -0.72
     immortal
    -0.71
    hattan
    -0.71
    POSITIVE LOGITS
     improperly
    1.29
     inadequate
    1.28
     deficiencies
    1.27
     inappropriately
    1.21
     improper
    1.17
     mishand
    1.17
     inadequ
    1.17
     inconsistent
    1.16
    adequ
    1.14
     negligent
    1.12
    Activations Density 0.826%

    No Known Activations