© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Natural Language
    Autoencoders
    NEW    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    3. Joseph Bloom · Feature Splitting for GPT2-Small
    5. GPT2-Small
    7. Residual Feature Splitting 49152
    9. 8-RES_FS49152-JB
    11. 45513
    Prev
    Next
    INDEX
    Explanations

    instances of the verb "was" and its derivatives in different contexts

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_fs49152-jb
    Configuration
    jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_49152
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    49,152
    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
    102
    +0.17
    0.8%
    55
    +0.14
    0.7%
    297
    +0.13
    0.6%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    102
    +0.17
    0.08
    55
    +0.14
    0.08
    566
    +0.13
    0.06
    Negative Logits
     caveats
    -0.76
     olds
    -0.73
     Extend
    -0.71
     depends
    -0.71
     Countries
    -0.71
     Preferences
    -0.70
     offsets
    -0.70
     IMAGES
    -0.69
     entails
    -0.69
     Strategies
    -0.68
    POSITIVE LOGITS
     murdered
    1.18
     assassinated
    1.16
     subjected
    1.15
     assaulted
    1.15
     stabbed
    1.10
     kidnapped
    1.09
     abducted
    1.08
     arrested
    1.07
     detained
    1.06
     harassed
    1.06
    Activations Density 0.214%

    No Known Activations