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. Residual Post - 32k
    5. 6-RES_POST_32K-OAI
    6. 4202
    Prev
    Next
    INDEX
    Explanations

    references to armed groups or conflicts

    oai_token-act-pair · gpt-4o-miniTriggered by @bot

    The neuron fires for news-agency style photo captions that narrate a subject’s present-tense action along with place and date details.

    oai_token-act-pair · o3Triggered by @kitfrasertaliente

    images or photo captions with dates and locations in news articles.

    oai_token-act-pair · claude-3-7-sonnet-20250219Triggered by @kitfrasertaliente

    references to specific people, locations, and dates within news report contexts.

    oai_token-act-pair · deepseek-r1Triggered by @kitfrasertaliente

    3 – TOP_POSITIVE_LOGITS mostly photography terms → photo captions

    np_max-act-logits · o3Triggered by @kitfrasertaliente
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 6-res_post_32k-oai
    Configuration
    jbloom/GPT2-Small-OAI-v5-32k-resid-post-SAEs/v5_32k_layer_6.pt
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    32,768
    Data Type
    torch.float32
    Hook Name
    blocks.6.hook_resid_post
    Hook Layer
    6
    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
    292
    +0.11
    0.5%
    642
    +0.11
    0.5%
    135
    +0.10
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    478
    +0.11
    0.05
    764
    +0.11
    0.04
    412
    +0.10
    0.04
    Negative Logits
    ©¶æ¥µ
    -0.81
    76561
    -0.64
    UCT
    -0.63
     hadn
    -0.62
    Need
    -0.62
    éŃĶ
    -0.62
    Ͻ
    -0.61
    Story
    -0.60
    phas
    -0.60
    cox
    -0.60
    POSITIVE LOGITS
     gestures
    0.93
    window
    0.86
     Photographer
    0.83
     poses
    0.83
     pose
    0.80
     celebrate
    0.76
     reacts
    0.73
     Photograph
    0.72
     IMAGES
    0.71
     maize
    0.71
    Activations Density 2.344%

    No Known Activations