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

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExportsSlackBlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Joseph Bloom · Feature Splitting for GPT2-Small
    3. GPT2-Small
    4. Residual Feature Splitting 1536
    5. 8-RES_FS1536-JB
    6. 533
    Prev
    Next
    INDEX
    Explanations

    Some of the activations were filtered out in the input, but based on the remaining activations, the neuron seems to be looking for numbers or labels related to different sections or topics within a text

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_fs1536-jb
    Configuration
    jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_1536
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    1,536
    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
    746
    +0.12
    0.5%
    336
    +0.12
    0.5%
    659
    +0.10
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    659
    +0.12
    0.12
    364
    +0.12
    0.08
    746
    +0.10
    0.06
    Negative Logits
     upside
    -0.64
    rosse
    -0.62
     unsupported
    -0.62
     Cerberus
    -0.60
     Vaj
    -0.59
     surve
    -0.59
     Izan
    -0.59
    draw
    -0.59
     roy
    -0.56
     inund
    -0.55
    POSITIVE LOGITS
    uggets
    1.28
    onsense
    1.28
    guyen
    1.23
    ucle
    1.21
    umerous
    1.20
    umeric
    1.14
    ihil
    1.12
    ominated
    1.11
    aughty
    1.10
    aturally
    1.08
    Activations Density 1.433%

    No Known Activations