Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsExportsSlackBlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Joseph Bloom · Open Source Sparse Autoencoders for all Residual Stream Layers of GPT2-Small
    3. GPT2-Small
    4. Residual Stream
    5. 0-RES-JB
    6. 7068
    Prev
    Next
    INDEX
    Explanations

    phrases instructing to maintain awareness or attention

    oai_token-act-pair · gpt-3.5-turbo

    the repeated phrase "Keep" in various contexts

    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
    »
    -0.67
     sporting
    -0.65
     hypot
    -0.65
    erson
    -0.64
    nes
    -0.63
     vacancy
    -0.63
    abilities
    -0.63
    azo
    -0.62
     paras
    -0.61
    ability
    -0.60
    POSITIVE LOGITS
     Keep
    3.44
    Keep
    2.32
    keep
    2.09
     Keeping
    1.55
     keep
    1.48
     Stay
    1.45
     Hold
    1.41
     Remember
    1.36
    Keeping
    1.30
     Protect
    1.30
    Activations Density 0.016%

    No Known Activations