© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Natural Language
    Autoencoders
    NEW    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    3. Gemma-2-27B
    5. 34-GEMMASCOPE-RES-131K
    7. 98847
    Prev
    Next
    INDEX
    Explanations

    gradients and gradient descent

    np_acts-logits-general · gemini-2.5-flash-lite

    The neuron strongly activates on mentions of gradient computation in neural‐network training (e.g. “gradient,” “descent,” “Rnn,” “training,” etc.).

    oai_token-act-pair · o4-miniTriggered by @jyhe0408
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-27b-pt-res/layer_34/width_131k
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    Resol
    -0.84
     Gohan
    -0.77
     Rhodes
    -0.75
    隐私
    -0.75
    ibles
    -0.75
    Rhodes
    -0.74
     belliger
    -0.73
     Assyrian
    -0.73
     Emmett
    -0.73
     Enzo
    -0.72
    POSITIVE LOGITS
     gradient
    1.54
     clipping
    1.35
     gradients
    1.31
     clip
    1.22
    grad
    1.20
     Gradient
    1.19
    gradient
    1.19
    Gradient
    1.18
     grad
    1.14
     exploding
    1.13
    Activations Density 0.011%

    No Known Activations