© 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. 20-GEMMASCOPE-2-TRANSCODER-262K
    4. 211624
    Prev
    Next
    INDEX
    Explanations

    The MAX_ACTIVATING_TOKENS list contains punctuation like `;`, `)`, and `{`, along with the keyword `import`. The TOKENS_AFTER_MAX_ACTIVATING_TOKEN list includes `import` and `sol`. The TOP_ACTIVATING_TEXTS show fragments of code snippets, including `import org.apache.camel...`, `import org.bukkit...`, `import net.dv8tion.jda...`, `import org.eclipse.swt...`, `import io.netty.channel...`, `import android.content...`, `import org.bukkit.plugin...`, and smart contract snippets like `Ownable.sol";` and `IERC721Receiver.sol";`.The common theme across these lists is code, programming languages, and software development. The punctuation often signifies code structure, `import` is a keyword for bringing in code libraries, and `.sol` refers to Solidity, a smart contract language.The TOP_POSITIVE_LOGITS list is quite varied and doesn't directly point to a single programming concept. However, `sol` in the `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` list strongly suggests smart contracts. Given the other code-related tokens, the neuron seems to activate when encountering code, particularly with structures involving imports or specific file types.Considering the instruction to be specific and find a pattern, the presence of `import` and `.sol` in contexts that are clearly code-related, and the general code snippets, points towards programming, specifically mentioning smart contracts due to `.sol`.Let's try to synthesize this. The neuron is activated by code structures. The presence of `import` and `sol` file extensions in code snippets implies a focus on programming elements.Possible explanations:- code imports- programming code- smart contract code- import statements- code and syntaxThe prompt asks for "what the neuron detects or predicts by finding patterns in lists".The pattern is clearly code. Specifically, given `sol` and `import`, it's related to programming languages that use import statements and potentially smart contract languages.The most salient combination is code syntax and libraries.Let's re-evaluate the lists with specific patterns in mind:- `MAX_ACTIVATING_TOKENS`: `;`, `)`, `{`, `import`, `.` - these are structural elements in code.- `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `import`, `sol` - keywords/file extensions in code.- `TOP_ACTIVATING_TEXTS`: Contains many `import` statements and `.sol` file references.The neuron seems to be identifying code segments, especially those involving library imports or specific scripting languages like Solidity.code structure and imports

    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_20_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
    ूब
    0.47
    Sans
    0.44
    هاز
    0.42
     sans
    0.42
     snee
    0.41
    🐝
    0.40
     bez
    0.40
     без
    0.39
    আবার
    0.39
     wher
    0.39
    POSITIVE LOGITS
    itate
    0.46
    ractic
    0.44
    akata
    0.42
    tev
    0.42
    pset
    0.42
     নেতা
    0.41
     oleh
    0.41
    shi
    0.41
     पत्थर
    0.40
    stered
    0.40
    Activations Density 0.000%

    No Known Activations