© 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. 40-GEMMASCOPE-2-RES-262K
    4. 132696
    Prev
    Next
    INDEX
    Explanations

    code blocks and syntax examples (```html, ```scala, ```python, ```elixir, ```javascript, ```twig, ```bash, ```jsx, ```groovy)2. Punctuation often found in code or definitions (`,`, `:`, `)`, `(`)3. Phrases indicating examples or explanations of code/syntax (`function`, `if`, `def`, `import`, `component`)4. Elements like `()`, `{}`, `[]` within code.The `TOP_POSITIVE_LOGITS` also show punctuation (`,…`, `...)`, `.…`, `(...`, `...`, `“)`), which strongly supports the idea of code or structured text.The most consistent pattern seems to be related to code, syntax, or structured text."code or structured text syntax" - 5 words. Fits the criteria."syntax examples and code" - 4 words. Fits the criteria."code blocks and syntax" - 4 words. Fits the criteria."code structure and syntax" - 4 words. Fits the criteria.Given the overall context from the examples and the logits, "code blocks and syntax" feels most descriptive. The `TOP_POSITIVE_LOGITS` also contain `loci` (places, often in code context like variable locations) and `either` (common in conditional logic). The `ളാ` is an outlier and might be noise or a very specific, less frequent pattern.Let's stick with a clear, common theme.Output: code blocks and syntax code blocks and syntax

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-it/resid_post/layer_40_width_262k_l0_medium
    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
     Voici
    0.52
    Voici
    0.50
    以下の
    0.47
     निम्नलिखित
    0.46
     cdot
    0.45
     Paste
    0.44
    下图
    0.44
     voici
    0.43
     permas
    0.43
     Рассмотрим
    0.43
    POSITIVE LOGITS
    ,…
    0.44
     ...)
    0.40
    .…
    0.39
    (...
    0.39
     ...
    0.39
    loci
    0.39
    ളാ
    0.38
    either
    0.38
    “)
    0.37
    threat
    0.37
    Activations Density 0.015%

    No Known Activations