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

    Neuronpedia

    Jacobian LensNEW
    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. 8-RES-JB
    11. 1013
    Prev
    Next
    INDEX
    Explanations

    conjunctions

    oai_token-act-pair · gpt-3.5-turbo

    document delimiters or markers indicating the end of a segment

    oai_token-act-pair · gpt-4o-miniTriggered by @bot

    Fires on ' and' or 'and' token

    Explanation Uploaded by User
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res-jb
    Configuration
    jbloom/GPT2-Small-SAEs-Reformatted/blocks.8.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.8.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    Skylion007/openwebtext
    Hook Point Layer
    8
    Activation Function
    relu
    Embeds
    IFrame
    Link
    •single token firings
    Fires on " and" (and if you put " and and and and" then it decreases in strength as you go)

    No Comments

    Negative Logits
    bub
    -0.66
    egu
    -0.62
    sylv
    -0.58
     distingu
    -0.54
    arching
    -0.52
    æ©
    -0.52
     Azerb
    -0.51
     Vaugh
    -0.50
    abul
    -0.49
    boarding
    -0.49
    POSITIVE LOGITS
    romeda
    0.89
    rew
    0.89
    rogens
    0.84
    rogen
    0.84
    ERSON
    0.82
    then
    0.72
    rea
    0.68
    RO
    0.66
     thence
    0.65
    rost
    0.63
    Activations Density 0.072%

    No Known Activations