© 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
    2. Gemma-3-27B-IT
    3. 31-GEMMASCOPE-2-TRANSCODER-262K
    4. 165059
    Prev
    Next
    INDEX
    Explanations

    The core task is to find a concise explanation (3-20 words) for a neuron's behavior based on the provided lists, focusing on patterns.1. **Analyze `MAX_ACTIVATING_TOKENS`**: These are tokens that strongly activate the neuron. * `the` * `<start_of_turn>` * `a` * `'` * `eg` * `’` * `urb` * `eg` (repeated)2. **Analyze `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`**: These are tokens immediately following the `MAX_ACTIVATING_TOKENS` that trigger strong activation. * `chat` (after `the`) * `2` (after `<start_of_turn>`) * `model` (after `a`) * `feminist` (after `'`) * `s` (after `eg`) * `le` (after `'`) * `s` (after `’`) * `s` (after `urb`) * `ate` (after `eg`) * `le` (after `eg`)3. **Analyze `TOP_POSITIVE_LOGITS`**: These are words/tokens the neuron associates with a high positive score.Based

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-it/transcoder_all/layer_31_width_262k_l0_small_affine
    Prompts (Dashboard)
    238,145 prompts, 512 tokens each
    Dataset (Dashboard)
    lmsys + oasst1
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    वु
    0.36
     Marshal
    0.35
     Tele
    0.35
     Survive
    0.34
    罗马
    0.34
     telephony
    0.34
     Emergency
    0.34
    远程
    0.34
     Ceremon
    0.34
     Avalanche
    0.33
    POSITIVE LOGITS
    erro
    0.38
    imos
    0.35
     méthod
    0.35
    enswert
    0.35
     الوحده
    0.34
    creation
    0.34
     들어가
    0.34
    Exclusive
    0.34
     doba
    0.34
     prejud
    0.33
    Activations Density 0.003%

    No Known Activations