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

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsBlogSlackPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Joseph Bloom · Feature Splitting for GPT2-Small
    3. GPT2-Small
    4. Residual Feature Splitting 6144
    5. 8-RES_FS6144-JB
    6. 4518
    Prev
    Next
    INDEX
    Explanations

    Commas used to introduce additional information, often following a conjunction or preceding a quote, particularly in complex sentences discussing social, political, or cultural topics.

    eleuther_acts_top20 · claude-3-5-sonnet-20240620Triggered by @jake

    Explanation could not be parsed.

    eleuther_acts_top20 · o3-miniTriggered by @jake

    Explanation could not be parsed.

    eleuther_acts_top20 · deepseek-r1Triggered by @jake

    The token "," before "because" in sentences, connecting clauses or expanding reasoning.

    eleuther_acts_top20 · deepseek-v3Triggered by @jake

    sentences containing em dashes (—) or hyphens (-) separating clauses or phrases, indicating pauses or additional information.

    oai_token-act-pair · deepseek-v3Triggered by @jake

    descriptions related to criminal activities and law enforcement

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_fs6144-jb
    Configuration
    jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_6144
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    6,144
    Data Type
    torch.float32
    Hook Point
    blocks.8.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    Skylion007/openwebtext
    Hook Point Layer
    8
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    720
    +0.14
    0.6%
    217
    +0.13
    0.6%
    498
    +0.11
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    217
    +0.14
    0.07
    720
    +0.13
    0.08
    498
    +0.11
    0.07
    Negative Logits
    olves
    -0.81
    SourceFile
    -0.73
    aukee
    -0.68
    zing
    -0.66
     Coverage
    -0.64
    ÅĤ
    -0.63
    nexus
    -0.62
    igers
    -0.60
    externalActionCode
    -0.59
    cium
    -0.59
    POSITIVE LOGITS
     alas
    1.20
     interestingly
    0.98
     moreover
    0.96
     unlike
    0.96
     unsurprisingly
    0.96
     contrary
    0.94
     although
    0.94
     according
    0.92
     despite
    0.92
     unfortunately
    0.90
    Activations Density 0.314%

    No Known Activations