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

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsExportsBlogSlackPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. OpenAI · Sparse Autoencoder for GPT2-Small - v5 2024
    3. GPT2-Small
    4. Attention Out - 128k
    5. 0-ATT_128K-OAI
    6. 13594
    Prev
    Next
    INDEX
    Explanations

    the word "block" and its variations, indicating a focus on obstacles or restrictions

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    jbloom/GPT2-Small-OAI-v5-128k-attn-out-SAEs/v5_128k_layer_0
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    131,072
    Data Type
    torch.float32
    Hook Name
    blocks.0.hook_attn_out
    Hook Layer
    0
    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
    216
    +0.20
    1.5%
    160
    +0.15
    1.2%
    336
    +0.14
    1.1%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    35
    +0.20
    0.03
    78
    +0.15
    0.03
    26
    +0.14
    0.03
    Negative Logits
    hovah
    -1.10
    ¿½
    -0.93
     fixme
    -0.93
    ensional
    -0.90
    ocument
    -0.87
    ktop
    -0.83
     incent
    -0.82
    monton
    -0.82
     Cheong
    -0.80
    ONSORED
    -0.76
    POSITIVE LOGITS
    iencies
    0.87
    worms
    0.80
    tered
    0.77
     Andromeda
    0.76
    Minecraft
    0.73
    abilities
    0.72
    picking
    0.72
    abases
    0.71
    ascript
    0.69
     darkness
    0.69
    Activations Density 0.035%

    No Known Activations