Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsExportsSlackBlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Joseph Bloom · Feature Splitting for GPT2-Small
    3. GPT2-Small
    4. Residual Feature Splitting 768
    5. 8-RES_FS768-JB
    6. 577
    Prev
    Next
    INDEX
    Explanations

    phrases related to ensuring specific conditions are met or actions are undertaken

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_fs768-jb
    Configuration
    jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_768
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    768
    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
    665
    +0.12
    0.5%
    133
    +0.10
    0.5%
    369
    +0.10
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    665
    +0.12
    0.23
    608
    +0.10
    0.17
    567
    +0.10
    0.09
    Negative Logits
    minus
    -0.71
     weeping
    -0.71
    option
    -0.68
     Warning
    -0.68
     Torn
    -0.68
     Verse
    -0.66
    sbm
    -0.66
    wrote
    -0.63
     Cosponsors
    -0.63
    elta
    -0.60
    POSITIVE LOGITS
     maximize
    1.25
     minimize
    1.15
     ensure
    1.15
     prevent
    1.09
     alleviate
    1.07
     improve
    1.07
     conserve
    1.07
     optimize
    1.07
     regain
    1.04
     preserve
    1.03
    Activations Density 6.005%

    No Known Activations