© 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. 37-GEMMASCOPE-2-TRANSCODER-262K
    4. 126712
    Prev
    Next
    INDEX
    Explanations

    **Neuron Explanation**The analysis of `MAX_ACTIVATING_TOKENS` and `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` shows a strong pattern of "message" followed by ")". This, combined with the `TOP_ACTIVATING_TEXTS` all containing `logging.basicConfig` and the `%(message)s` format string, points towards the neuron activating on log messages.Therefore, the neuron detects or predicts content related to log messages, specifically their structure or presence within formatted strings.Considering the rules:- Concise (3-20 words).- Find patterns.- Not listing all tokens.- No "tokens", "patterns", "detects", "predicts".- Specific.The most prominent pattern is anything related to `logging.basicConfig(..., format='...%(message)s')`. The core output being identified is the `message` part of a log.Final answer should be a short phrase.log message contentsformatted logging messageslogging message contentcontent in log messages"formatted logging messages" seems to capture the essence well, noting that the `format='%(asctime)s - %(levelname)s - %(message)s'` is key, and the neuron is likely activating on patterns within those formatted messages.Let's consider just the specific pattern found from MAX_ACTIVATING_TOKENS and TOKENS_AFTER_MAX_ACTIVATING_TOKEN: "message)". This is part of the format string itself often. The neuron might be looking for this specific structure within log messages.The `TOP

    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_37_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
     हनी
    0.44
     CCP
    0.38
     पश्चात
    0.38
     Researcher
    0.38
     AFTER
    0.37
    ACITY
    0.37
     MODEL
    0.37
     Moroccan
    0.37
     STUDY
    0.36
    ijima
    0.36
    POSITIVE LOGITS
     iphdr
    0.39
     throne
    0.38
    legraph
    0.37
    Prem
    0.37
    graphs
    0.36
    ナン
    0.36
    Graphs
    0.36
     Davison
    0.35
    isão
    0.35
    ਪ
    0.35
    Activations Density 0.000%

    No Known Activations