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 24576
    5. 8-RES_FS24576-JB
    6. 19301
    Prev
    Next
    INDEX
    Explanations

    phrases related to instructions or calls to action

    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
    601
    +0.12
    0.5%
    377
    +0.10
    0.5%
    212
    +0.10
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    566
    +0.12
    0.05
    762
    +0.10
    0.06
    92
    +0.10
    0.03
    Negative Logits
     retrieval
    -0.70
    catentry
    -0.64
    mentioned
    -0.62
    umbnails
    -0.62
     Citation
    -0.61
    course
    -0.61
     coroner
    -0.59
    ibaba
    -0.59
     strand
    -0.59
     Medline
    -0.59
    POSITIVE LOGITS
     exist
    0.89
     trustworthy
    0.75
     cared
    0.75
     cares
    0.74
     sorry
    0.73
     exists
    0.72
    chwitz
    0.72
     okay
    0.71
     belong
    0.71
     messed
    0.71
    Activations Density 0.293%

    No Known Activations