© 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. 3-RES-JB
    11. 20901
    Prev
    Next
    INDEX
    Explanations

    the word "dogged" and similar terms related to persistent challenges or obstacles

    oai_token-act-pair · gpt-3.5-turbo

    terms related to ongoing struggles or difficulties

    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
     sidx
    -0.66
    reme
    -0.66
    pressed
    -0.64
    endi
    -0.63
    RN
    -0.63
    hal
    -0.62
    unicip
    -0.62
    achers
    -0.61
    yl
    -0.61
    raviolet
    -0.60
    POSITIVE LOGITS
     plagued
    0.95
     dogged
    0.87
     plag
    0.84
    tenance
    0.82
     Ples
    0.74
     overshadowed
    0.73
    luster
    0.71
    ousel
    0.69
     ..........
    0.67
    DAQ
    0.67
    Activations Density 0.030%

    No Known Activations