© 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 · Open Source Sparse Autoencoders for all Residual Stream Layers of GPT2-Small
    5. GPT2-Small
    7. Residual Stream
    9. 3-RES-JB
    11. 21158
    Prev
    Next
    INDEX
    Explanations

    words related to violent physical actions, particularly involving destruction

    oai_token-act-pair · gpt-3.5-turbo

    instances of the word "smash" and its variations, indicating destructive actions

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 3-res-jb
    Configuration
    jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.hook_resid_pre
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    24,576
    Data Type
    torch.float32
    Hook Point
    blocks.3.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    Skylion007/openwebtext
    Hook Point Layer
    3
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    ghan
    -0.69
    anamo
    -0.69
    abet
    -0.68
    arij
    -0.65
     Senators
    -0.64
    isan
    -0.64
    annis
    -0.63
    rollment
    -0.62
    ople
    -0.61
    itent
    -0.61
    POSITIVE LOGITS
     smashed
    0.92
     smash
    0.92
     smashing
    0.81
    ulence
    0.80
    ãĥ¼ãĥ«
    0.73
    ulously
    0.73
     shattered
    0.71
    adoes
    0.71
     shattering
    0.70
     down
    0.69
    Activations Density 0.012%

    No Known Activations