© 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. 15-GEMMASCOPE-2-TRANSCODER-262K
    4. 14651
    Prev
    Next
    INDEX
    Explanations

    This neuron seems to be associating phrases that describe finishing an action, task, or process. The `MAX_ACTIVATING_TOKENS` often contain words like "with", "execution", "work", "on", and "silence" which can indicate completion or a state after an action. The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` gives further clues: "authoritarian" (rule, end of an era), "delivering" (completing a service), "feeling" (experiencing an outcome), "heart" (core of something after processing), "smiles" (outcome of satisfaction).The `TOP_POSITIVE_LOGITS` include words in various languages, suggesting a broad semantic scope for completion or finality, but without a clear English semantic link. However, the `TOP_ACTIVATING_TEXTS` is where the pattern becomes clearer:- "proceeded with the procedure" (completing a medical task)- "resumes execution *from where it left off*" (completing a step, continuing)- "beginning his own decades-long authoritarian rule" (end of one era, start of another)- "focus on what you do best – delivering quality products and experiences" (completing a service)- "focus on the logic of your application" (completing a setup)The common thread is completing a task, bringing something to a conclusion, or the state of being after a process. Looking for a concise phrase.Keywords: `with`, `execution`, `delivering`, `rule`, `logic`, `procedure`.Seems to be about finishing, completing, or concluding an action or state.Let's try to formulate a phrase."completing actions or states" - 4 words, fits."finishing tasks and processes" - 4 words, fits."describes task completion" - 3 words, fits."phase transitions or conclusions

    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_15_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
    IN
    0.18
    >')
    0.18
     strain
    0.17
    ス
    0.17
    INTERNAL
    0.17
     шум
    0.16
     propon
    0.16
    냈
    0.16
    াশ
    0.16
    ів
    0.16
    POSITIVE LOGITS
    valids
    0.19
     Ró
    0.17
    니다
    0.17
     ยาก
    0.16
    arabic
    0.16
     étrangères
    0.16
     භ
    0.16
     langu
    0.16
     Breakfast
    0.16
     没有
    0.16
    Activations Density 0.103%

    No Known Activations