© 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-4-31B
    3. 30-RES-MATRYOSHKA-131K
    4. 130172
    Prev
    Next
    INDEX
    Explanations

    class DoubleWrap{ public: static int id; DoubleWrap(double value){this->value = value;myid=++id;} double value; int myid; std::Here, 'value' is a member variable within a class definition, or a parameter in a constructor. The tokens after `value` are `)` or `.`, or it's part of a variable assignment like `this->value = value;`.Let's look for this pattern.In `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`, I see:- `is` (e.g., `self.is_create=False)`)- `transform` (e.g., `self.transform = transform`)- `log` (e.g., `self._logfn = logfn`)- `context` (e.g., `this.context = context`)- `visual` (e.g., `this.visualizer = visualizer`)- `app` (e.g., `this.app = app`)- `name` (e.g., `self.name = name`)- `f` (e.g., `self.f_name = f_name`)- `dom` (e.g., `this.dom = dom`)- `=` (e.g., `self.is_create=False)`)These seem to be assignments or initializations within class constructors or methods, where `value` is likely a parameter or a member variable being assigned.The phrase structure is often `self.variable = parameter` or `this.variable = parameter`, or `variable(value)`.The `MAX_ACTIVATING_TOKENS` is just `value`.The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` shows assignments or parameter usage.The `TOP_POSITIVE_LOGITS` are general English words, some of which are less common or might indicate a broader context, but not directly helpful for pinpointing a specific token pattern.The recurring pattern in `TOP_ACTIVATING_TEXTS` is the structure of object-oriented programming, often involving `class`, `__init__`, `constructor`, and assignments like `self.name = name`, `this.context = context`, `self.value = value`.class variable initialization

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    decoderesearch/gemma-4-saes/gemma-4-31b
    Prompts (Dashboard)
    16,384 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    কিন্তু
    -0.06
    重视
    -0.06
     však
    -0.06
    醪
    -0.06
    方に
    -0.06
    ܬ
    -0.06
     tendrá
    -0.06
     will
    -0.06
    何を
    -0.06
     hingegen
    -0.06
    POSITIVE LOGITS
     through
    0.06
    ثل
    0.06
     specially
    0.06
    udin
    0.05
    bounds
    0.05
     pouvoir
    0.05
     accidentally
    0.05
     when
    0.05
     இணைய
    0.05
    through
    0.05
    Activations Density 0.003%

    No Known Activations