© 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. 0-RES-JB
    11. 19026
    Prev
    Next
    INDEX
    Explanations

    mentions of the word "witch" followed by different characters or words occurring in different contexts

    oai_token-act-pair · gpt-3.5-turbo

    words related to witchcraft and types of products or entities

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 0-res-jb
    Configuration
    jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    Skylion007/openwebtext
    Hook Point Layer
    0
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     Sandy
    -0.67
     Tart
    -0.64
    sand
    -0.64
     upkeep
    -0.63
     Ike
    -0.63
     Leban
    -0.63
     coping
    -0.62
     Ley
    -0.61
     Arlington
    -0.60
     Jiu
    -0.58
    POSITIVE LOGITS
    usp
    2.30
    ype
    2.20
    witch
    2.03
     Marqu
    1.80
    ocus
    1.40
    ypes
    1.30
    actory
    1.18
    mid
    1.15
    ynt
    0.99
     marqu
    0.96
    Activations Density 0.046%

    No Known Activations