© 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 EvalsExportsinterp-engineNEWAPI 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. 16147
    Prev
    Next
    INDEX
    Explanations

    quotations or reported speech

    oai_token-act-pair · gpt-3.5-turbo

    sentences that contain a closing quotation mark

    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
     skelet
    -0.76
     pit
    -0.71
     tremend
    -0.67
    kw
    -0.66
    byss
    -0.66
     suspic
    -0.65
    rieved
    -0.64
     slam
    -0.62
     prelim
    -0.61
     boarding
    -0.60
    POSITIVE LOGITS
     Whilst
    0.87
     Keynes
    0.84
    ablishment
    0.76
     Alternatively
    0.75
     Firstly
    0.74
     Seems
    0.74
    Firstly
    0.73
     Presumably
    0.71
     Depending
    0.71
    arta
    0.70
    Activations Density 0.034%

    No Known Activations