© 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. OpenAI · Sparse Autoencoder for GPT2-Small - v5 2024
    5. GPT2-Small
    7. Residual Post - 32k
    9. 8-RES_POST_32K-OAI
    11. 6329
    Prev
    Next
    INDEX
    Explanations

    technical terms and phrases related to software performance and design flaws

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_post_32k-oai
    Configuration
    jbloom/GPT2-Small-OAI-v5-32k-resid-post-SAEs/v5_32k_layer_8.pt
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    32,768
    Data Type
    torch.float32
    Hook Name
    blocks.8.hook_resid_post
    Hook Layer
    8
    Architecture
    standard
    Context Size
    64
    Dataset
    Skylion007/openwebtext
    Activation Function
    topk
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    290
    +0.12
    0.5%
    260
    +0.11
    0.5%
    633
    +0.10
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    635
    +0.12
    0.09
    762
    +0.11
    0.21
    282
    +0.10
    0.16
    Negative Logits
     congr
    -0.69
     congratulations
    -0.67
    knit
    -0.65
    assador
    -0.63
     GOODMAN
    -0.63
    romeda
    -0.60
     dign
    -0.60
    reetings
    -0.59
    psons
    -0.59
     congratulated
    -0.57
    POSITIVE LOGITS
     inadequate
    1.39
     poorly
    1.37
     incorrectly
    1.37
     inaccurate
    1.36
     inconsist
    1.35
     inconsistent
    1.32
    failed
    1.32
     insufficient
    1.30
     ineffective
    1.30
     incorrect
    1.30
    Activations Density 13.677%

    No Known Activations