© 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. 15-GEMMASCOPE-2-TRANSCODER-262K
    4. 16398
    Prev
    Next
    INDEX
    Explanations

    patterns related to counting, asking, and independent entities.This neuron appears to focus on actions related to counting and asking questions, often in the context of independence or objective facts.Let's break down the lists to find a unifying theme:* **MAX_ACTIVATING_TOKENS**: `, count, ask, call, вати, define, ask, independent, counting, you` * Key verbs/concepts: `count`, `ask`, `define`, `independent`, `counting`. * The presence of Russian (`вати`) and Ukrainian (`розпаду Радянського Союзу.`).* **TOKENS_AFTER_MAX_ACTIVATING_TOKEN**: `China, from, !, it, від, ", ., states, ., '` * These seem like punctuation or common following words, but "China" and "states" appear.* **TOP_POSITIVE_LOGITS**: `lec, म, ুম, Doctrine, McCon, මු, than, khấu, mys, cutter` * These are less clear. `Doctrine`, `McCon` might hint at specific entities or concepts. `lec` could be part of "lecture" or "selected". The other tokens are non-Latin script.* **TOP_ACTIVATING_TEXTS**: * Mentions "independence", "countries", "US", "EU", "Serbia", "Russia", "China", "Kosovo", "Serbs". This theme is strong.counting, asking "independent"

    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
     ovaj
    0.33
     attenzione
    0.32
    はなく
    0.32
    haben
    0.31
    eball
    0.31
    hoe
    0.30
    hatian
    0.30
     அன்பு
    0.30
    randomNumber
    0.30
     potentialities
    0.29
    POSITIVE LOGITS
     lec
    0.34
     म
    0.33
    ুম
    0.33
    Doctrine
    0.33
     McCon
    0.33
     මු
    0.33
     than
    0.33
     khấu
    0.33
    mys
    0.33
     cutter
    0.32
    Activations Density 0.086%

    No Known Activations