© 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-3-12B-IT
    3. 12-GEMMASCOPE-2-RES-262K
    4. 2840
    Prev
    Next
    INDEX
    Explanations

    " * "@app.get("/items/{item_id}", response_model=Item) async def read_item(item_id: int):" - Contains `/items/{item_id}`. * "return todos @app.get("/todos/{todo_id}", response_model=Todo) async def read_todo(todo_id: int):" - Contains `/todos/{todo_id}`. * "1. **If it returns a single specification:** * /api/v1/specifications/{specificationId}`" - Contains `/specifications/{specificationId}`. * "paths: /users/{userId}: get: summary: Returns a user by ID. parameters: - name" - Contains `/users/{userId}`. * "2. Get a specific todo by ID (GET /todos/<todo_id>)" - Contains `/todos/<todo_id>`. * "You requested user with ID: ${userId}); }); // Route with multiple URL parameters app.get('/products/:category/:productName'," - Contains `/products/:category/:productName`. * "* **URI:** `https://graph.microsoft.com/v1.0/sites/{SiteURL}/usage`" - Contains `{SiteURL}/` which is a placeholder for a site identifier. * "* **Layouts:** Use the `app/layout.tsx` and `app/[path]/layout.tsx` to define the overall structure of your application." - Contains `/[path]/`. * "then start it again, respecting any defined startup and shutdown sequences. **2. Using `/etc/init.d/<service_name>` directly" - Contains `/<service_name>`. * "Installation Directory:** The default location is often something like: * **Windows:** `C:\Users\<YourUsername>\miniconda3`" - Contains `\<YourUsername>\`.**Pattern Identification:**The `MAX_ACTIVATING_TOKENS` are `/{`, `/<`, `/:`, `/[`, `/\`. These are all path separators or delimiters commonly used in URLs and file paths.The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` are words like `item`, `todo`, `specification`, `userId`, `category`, `Site`, `path`, `service`. These are often the resource names or identifiers that follow these path separators in API endpoints or file system paths.The `TOP_ACTIVATING_TEXTS` overwhelmingly show examples of URLs and API endpoints that use these path separators followed by specific resource names (e.g., `/items/{item_id}`, `/users/{userId}`, `/todos/<todo_id>`, `/[path]/`).The `TOP_POSITIVE_LOGITS` list contains mostly non-English words. This might suggest that the neuron also has a secondary activation for non-English content, or perhaps these are less significant for the primary pattern. However, the overwhelming evidence points towards URL/path structures.**Conclusion:**The neuron clearly identifies structures common in defining resource paths and endpoints, especially those with parameterized segments indicated by curly braces `{}` or angle brackets `<>`.**Explanation:**resource paths like /users/{userId} or /items/{itemId}resource paths like /users/{userId} or /items/{itemId}

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-12b-it/resid_post/layer_12_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
    ak
    1.79
    ب
    1.72
    to
    1.58
    ya
    1.55
    ている
    1.54
    c
    1.52
    かつ
    1.46
    ਰ
    1.44
    ere
    1.43
    ab
    1.42
    POSITIVE LOGITS
     насеко
    1.44
     smě
    1.43
    ა
    1.42
     Владимира
    1.41
     каче
    1.40
     способности
    1.40
     Тре
    1.38
     peque
    1.34
    ।।
    1.34
     সময়ে
    1.34
    Activations Density 0.020%

    No Known Activations