© 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. 3-GEMMASCOPE-2-TRANSCODER-262K
    4. 46322
    Prev
    Next
    INDEX
    Explanations

    battery pack architecture and componentsHere's the thinking process:1. **Analyze `MAX_ACTIVATING_TOKENS`**: The most frequent token is "pack". "Backup" and "Powered" also appear. This immediately suggests a connection to "battery packs" or "backup power".2. **Analyze `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`**: * After "pack": "architecture", "such". * After "Backup"/"backup": "can", "to". This helps refine the "pack" idea. "pack architecture" is a strong pairing.3. **Analyze `TOP_POSITIVE_LOGITS`**: The presence of "électrique" (French for "electric") is a very strong signal that this neuron is related to electrical or battery systems. The numbers (5, 6, 7, 8, 9) are less informative on their own, but often appear in technical specifications or quantities.4. **Analyze `TOP_ACTIVATING_TEXTS`**: * "battery pack architecture and components" (appears multiple times) * "battery pack" followed by "components" or discussions about its function. * Other texts also mention "battery pack" in technical contexts (e.g., "80V Battery pack", "third module of the battery pack"). * There's also a mention of "Battery Backup" related to power outages.5. **Synthesize and Identify the Core Pattern**: * "pack" is central. * "architecture" frequently follows "pack". * "battery" is implied by the context and the "électrique" logit. * "components" is also strongly associated with "pack architecture". * "Backup" is a secondary theme, but "battery pack architecture" is the dominant pattern in the activating texts.6. **Formulate the Explanation (Concise, 3-20 words, no "tokens"/"patterns", not a full sentence)**: * "battery pack architecture" (4 words) - Good, but could be more specific if "components" is also key. * "battery pack architecture and components" (5 words) - This perfectly captures the most frequent and specific phrases found, especially from the activating texts. It's well within the word count. * "backup power" (2 words) - Too short and doesn't capture the "pack architecture" theme. * "electrical pack" (2 words) - Too generic.7. **Final Choice**: "battery pack architecture and components" is the most specific and representative phrase derived from the provided lists.battery pack architecture and components

    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_3_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
    i
    1.06
    o
    1.05
    n
    1.02
    a
    0.93
    an
    0.89
    น
    0.86
    in
    0.80
    ac
    0.78
    н
    0.73
    ن
    0.73
    POSITIVE LOGITS
    9
    0.79
    8
    0.79
     for
    0.75
    у
    0.71
     électrique
    0.68
    5
    0.68
    6
    0.68
    ية
    0.67
    7
    0.66
    𝕤
    0.65
    Activations Density 0.001%

    No Known Activations