© 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. 18088
    Prev
    Next
    INDEX
    Explanations

    **TOP_ACTIVATING_TEXTS**: * "...German deaths at Stalingrad..." * "...encircling enemy formations, cutting them off from supplies and reinforcements. German forces were equipped with..." * "Russian forces have been attempting to encircle the city... Ukrainian forces are putting up strong resistance." * "...harass Russian advances, conduct ambushes, and exploit weaknesses." * "...locating, and then suppressing/destroying enemy radar and SAM sites." * "...suffering from Union military campaigns. Moving the war to the North..." * "...hit Russian oil depot..." * "...counter the Russian attacks." * "...German lines in Normandy. It ultimately stalled with heavy casualties." * "...disable or disrupt enemy unmanned aerial systems (drones)."**Pattern Analysis:*** **MAX_ACTIVATING_TOKENS**: Consistently shows nationalities or sides in conflict ("German", "Russian", "enemy", "Union").* **TOKENS_AFTER_MAX_ACTIVATING_TOKEN**: Words directly related to war and military action ("deaths", "formations", "forces", "advances", "radar", "military", "attacks", "lines").* **TOP_ACTIVATING_TEXTS**: Provides numerous examples of specific military conflicts, battle outcomes, and military actions involving "German", "Russian", "Union", and "enemy" forces. The context is overwhelmingly military and conflict-related.**Synthesis:**The neuron seems to activate when encountering terms related to military conflict, specifically involving or referencing "German", "Russian", or "Union" forces, followed by actions or outcomes of war. The "TOP_POSITIVE_LOGITS" are less informative here, possibly indicating a multi-lingual or abstract component, but the dominant pattern is conflict.**Explanation:**warfare involving German or Russian forces

    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
    ruptcy
    0.42
     overarching
    0.39
    ματος
    0.37
     beak
    0.36
    inoc
    0.36
     poaching
    0.36
     dusting
    0.35
     rover
    0.35
    ruff
    0.35
     Latham
    0.35
    POSITIVE LOGITS
     Ocak
    0.41
     Balliye
    0.38
    uvian
    0.37
    k
    0.37
     Herald
    0.36
     Seguro
    0.36
     রাজনীতিবিদ
    0.36
    ﻠ
    0.36
     알
    0.36
    알
    0.35
    Activations Density 0.014%

    No Known Activations