© 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 EvalsExportsinterp-engineNEWAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 34-GEMMASCOPE-2-TRANSCODER-262K
    4. 167441
    Prev
    Next
    INDEX
    Explanations

    **Analyzing Neuron Input**I'm examining the provided lists to identify a consistent pattern that describes the neuron's function.* **MAX_ACTIVATING_TOKENS**: `frac`, `.`, `=`, `frac`, `frac`, `2`, `frac`, `6`, `**`* **TOKENS_AFTER_MAX_ACTIVATING_TOKEN**: `{`, `2`, `\`, `{`, `{`, `1`, `.`, `{`, `.`, `3`* **TOP_POSITIVE_LOGITS**: `략`, `icionados`, `ليس`, `Consul`, `striction`, `гія`, `视为`, `내용을`, `Issled`, `呫`* **TOP_ACTIVATING_TEXTS**: This list contains many examples of mathematical expressions, specifically ones involving fractions and equations. Examples include: * "`$1.609344 = \frac{1.609344}{1} = \frac{1609344}{}$`" * "`$ x = \frac{645}{0.25} $`" * "`$b = \frac{73.50}{21}$`" * "`$t_1 = \frac{6.75}{1.5}$`" * "`log2(50/6.25). We can simplify the fraction 50/6.25`" * "`3.25** or **3 1/4**`"**Pattern Identification:**The `MAX_ACTIVATING_TOKENS` list clearly shows a strong preference for `frac`.The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` list shows characters like `{`, `2`, `1`, `3`, suggesting that `frac` is often part of a structured notation, like `\frac{numerator}{denominator}`.The `TOP_ACTIVATING_TEXTS` provide definitive context. Many examples contain `\frac{}{} ` syntax or discuss fractions extensively. The `TOP_POSITIVE_LOGITS` are diverse and non-English, which might suggest the neuron is broadly interested in symbolic representations or mathematical contexts that appear across languages, but based on the other lists, the primary trigger is mathematical fraction notation.The most consistent and specific pattern is the occurrence and handling of mathematical fractions, often represented using LaTeX-like syntax (`\frac{}{} `).**Explanation:**

    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_34_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
    percent
    0.40
    enario
    0.39
     commonly
    0.38
    junior
    0.38
    commonly
    0.38
    ∕
    0.38
    overhead
    0.37
     multiv
    0.37
    .$-
    0.36
     overhead
    0.36
    POSITIVE LOGITS
    략
    0.38
    icionados
    0.37
     ليس
    0.37
     Consul
    0.37
    striction
    0.37
    гія
    0.37
    视为
    0.36
     내용을
    0.36
     Issled
    0.35
    呫
    0.35
    Activations Density 0.002%

    No Known Activations