© 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. 877
    Prev
    Next
    INDEX
    Explanations

    examples of specific information or facts related to events, equipment, or procedures

    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
    565
    +0.10
    0.5%
    654
    +0.10
    0.5%
    512
    +0.10
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    512
    +0.10
    0.07
    718
    +0.10
    0.08
    102
    +0.10
    0.08
    Negative Logits
     Saud
    -0.70
     Lies
    -0.69
     Secrets
    -0.68
     Lists
    -0.67
     Reporting
    -0.67
     Brow
    -0.65
     Claims
    -0.65
     Visitors
    -0.65
     Investigative
    -0.65
     Lauder
    -0.64
    POSITIVE LOGITS
     preferable
    1.61
     feasible
    1.30
     advisable
    1.28
     preferred
    1.20
     viable
    1.17
     impractical
    1.11
     desirable
    1.09
     advantageous
    1.06
     recommended
    1.06
     achievable
    1.04
    Activations Density 0.924%

    No Known Activations