© 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. 37-GEMMASCOPE-2-TRANSCODER-262K
    4. 136445
    Prev
    Next
    INDEX
    Explanations

    **Detecting Code Snippets**I have analyzed the provided lists. The `MAX_ACTIVATING_TOKENS` show `class` frequently appearing, and `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` lists common programming terms (like `Time`, `even`, `sum`, `String`, `count`, `array`, `add`). The `TOP_ACTIVATING_TEXTS` confirm this by showing examples of code blocks in languages like Java, Rust, Dart, Go, and C, often illustrating how to run or compile code. The `TOP_POSITIVE_LOGITS` list includes words like `multimedia`, which seems less related, but also `Answers` or `답변` (Korean for answer) suggesting providing solutions or examples.The core pattern is the neuron activating when it sees the `class` keyword in code, followed by potential identifiers related to programming tasks or concepts.Therefore, a concise explanation for the neuron's behavior is:**class keyword followed by code identifiers**

    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_37_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
    iguez
    0.39
    configuration
    0.36
    efficacy
    0.36
    sparse
    0.35
     tasa
    0.35
    conciliation
    0.35
    max
    0.34
    asc
    0.34
     канале
    0.34
    感恩
    0.34
    POSITIVE LOGITS
    პერ
    0.46
     Antworten
    0.41
     multimedia
    0.40
     답변
    0.40
     diterima
    0.39
     Myth
    0.37
    ০১
    0.37
     عبدال
    0.37
     vle
    0.36
    Multimedia
    0.36
    Activations Density 0.002%

    No Known Activations