© 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 49152
    9. 8-RES_FS49152-JB
    11. 41729
    Prev
    Next
    INDEX
    Explanations

    references to numerical rankings or levels of importance

    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
    261
    +0.10
    0.4%
    391
    +0.09
    0.4%
    260
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    555
    +0.10
    0.03
    260
    +0.09
    0.03
    396
    +0.09
    0.02
    Negative Logits
     [|
    -0.71
    Interstitial
    -0.67
     Horses
    -0.66
    TPPStreamerBot
    -0.66
    ajor
    -0.65
    ournals
    -0.64
     Syndrome
    -0.64
    lishes
    -0.62
     Mania
    -0.61
    Film
    -0.59
    POSITIVE LOGITS
     prosecutor
    1.01
     aide
    0.98
     diplomat
    0.97
    ranking
    0.96
    most
    0.95
     adviser
    0.87
     brass
    0.87
     scorer
    0.86
     leadership
    0.84
     aides
    0.83
    Activations Density 0.030%

    No Known Activations