© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Natural Language
    Autoencoders
    NEW
    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 16-GEMMASCOPE-2-TRANSCODER-262K
    4. 215998
    Prev
    Next
    INDEX
    Explanations

    Analyzing Neuron BehaviorI am currently focused on distilling the core mechanism of neuron activation from the provided input lists. My aim is to encapsulate this behavior into a brief, specific phrase, avoiding any unnecessary preamble.Looking at the `MAX_ACTIVATING_TOKENS`: `api`, `late`, `beginning`, `problems`.Looking at the `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `)` (after `api`), `of` (after `api`), `Capital` (after `late`), `raw` (after `beginning`), `for` (after `ake`), `and` (after `ed`), `?"` (after `problems`).Analyzing Neuron BehaviorI am currently focused on distilling the core mechanism of neuron activation from the provided input lists. My aim is to encapsulate this behavior into a brief, specific phrase, avoiding any unnecessary preamble.The `MAX_ACTIVATING_TOKENS` show words like `api`, `late`, `beginning`, `problems`.The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` often show punctuation or common words: `)`, `of`, `Capital`, `raw`, `for`, `and`, `?"`.`TOP_POSITIVE_LOGITS` contain diverse words in different scripts, not immediately forming a clear semantic pattern with the activating tokens.`TOP_ACTIVATING_TEXTS` include code snippets (`import win32gui`, `api`), descriptions of time (`late 20th-century`, `beginning of LLMs`), and problem descriptions (`creates problems`).The common thread seems to be relating to common programming/library terms (like `api`) and concepts of time (`late`, `beginning`) or issues (`problems`). The tokens after `api` (`of`, `)`) suggest it might be related to function calls or definitions. The token `Capital` after `late` and

    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_16_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
    4
    0.55
     doh
    0.54
    3
    0.52
    x
    0.48
    HON
    0.48
    6
    0.48
     Diplomacy
    0.48
     esk
    0.47
    חות
    0.47
     sklad
    0.47
    POSITIVE LOGITS
     गर्
    0.49
    PAssignment
    0.47
    alupe
    0.47
     کال
    0.46
     verpflichtet
    0.46
     نیست
    0.45
    ຮູ້
    0.45
     χρει
    0.45
     bolstered
    0.45
     ຮ
    0.44
    Activations Density 0.000%

    No Known Activations