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

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports 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. 6409
    Prev
    Next
    INDEX
    Explanations

    phrases related to connections or relationships

    oai_token-act-pair · gpt-3.5-turbo

    instances of the word "relate" and its variations

    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
    onew
    -0.69
    airs
    -0.69
    kers
    -0.67
    ifter
    -0.67
    iami
    -0.67
    avis
    -0.64
    ulsion
    -0.62
    apo
    -0.62
    aan
    -0.61
    odder
    -0.61
    POSITIVE LOGITS
     relate
    0.84
    worldly
    0.82
    MpServer
    0.74
    ability
    0.74
     thereto
    0.72
     positively
    0.71
    metry
    0.71
     relates
    0.70
     ancest
    0.69
     favorably
    0.68
    Activations Density 0.015%

    No Known Activations