© 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. 27-GEMMASCOPE-2-TRANSCODER-262K
    4. 196710
    Prev
    Next
    INDEX
    Explanations

    This neuron seems to be related to sentences describing actions or events, often initiated by a subject and followed by a description of what happened or what is being done.Looking at MAX_ACTIVATING_TOKENS and TOKENS_AFTER_MAX_ACTIVATING_TOKEN:- "I" followed by "then" (implied from "I then carry...")- "was" followed by context (e.g., "was waiting", "was informed")- "her" followed by "shopping" or "chatting"- "After" followed by "chatting"Looking at TOP_POSITIVE_LOGITS:'Marks', 'realizes', 'Properties', 'Marsh' could be names or concepts that appear in descriptive text.The phrases in TOP_ACTIVATING_TEXTS are descriptive narratives or explanations.The core theme seems to be about describing events, actions, or states.Let's try to find a simpler, more direct pattern.The MAX_ACTIVATING_TOKENS are often the subjects or start of clauses.The TOKENS_AFTER_MAX_ACTIVATING_TOKEN are often verbs or subsequent actions/objects.Consider the context: "I then carry...", "gentleman was informed", "her shopping list", "After chatting".It's about *what follows* the initial part of a statement or action.The TOP_POSITIVE_LOGITS lean towards names and concepts that might be part of such descriptions or scenarios.describing subsequent events

    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_27_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
     flickering
    0.27
     paddingTop
    0.26
     Welding
    0.26
     coupes
    0.26
     finalText
    0.25
     Cleanup
    0.25
     ST
    0.25
     Lighting
    0.25
     DOL
    0.25
     tempHeader
    0.25
    POSITIVE LOGITS
    ўні
    0.26
     mark
    0.25
    ítmény
    0.25
     estão
    0.25
     মার্শাল
    0.25
    Marks
    0.24
     realizes
    0.24
    Properties
    0.24
    ψη
    0.23
    Marsh
    0.23
    Activations Density 0.000%

    No Known Activations