© 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 EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 38-GEMMASCOPE-2-TRANSCODER-262K
    4. 4188
    Prev
    Next
    INDEX
    Explanations

    The presence of `annotation` as the `MAX_ACTIVATING_TOKENS` and its repetition, combined with the `TOP_ACTIVATING_TEXTS` showing many import statements from the Spring framework (like `org.springframework.scheduling.annotation.Scheduled`), strongly suggests this neuron is related to Java code annotations, particularly within the Spring ecosystem.The most fitting and concise explanation would capture this context.- "annotation" is the key item in `MAX_ACTIVATING_TOKENS`.- The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` being `.` is typical for code.- The `TOP_ACTIVATING_TEXTS` confirm this is about code imports and annotations.A good phrase could be "spring annotations" or "java annotations". "spring annotations" is more specific given the texts.Constraint Checklist & Confidence Score:1. Concise explanation (3-20 words): Yes. "spring annotations" is 2 words.2. Single word, phrase, or pattern: Yes. "spring annotations" is a phrase.3. About tokens following/preceding: The pattern is derived from this, but the output is not explicitly describing the relationship.4. About words starting with a sequence: Not directly, but the context implies specific types of words/code constructs.5. Avoid listing tokens: Yes.6. Just say the pattern itself: Yes.7. No "words related to", "concepts related to", etc.: Yes.8. No "This neuron detects/predicts": Yes.9. No "tokens" or "patterns": Yes.10. No capitalization unless proper noun: Yes. "spring" is lowercase, "annotations" is lowercase.11. Specific: Yes. "spring annotations" is specific.12. Majority match: Yes, the majority of activating texts are Spring annotations.13. If cannot guess, return first activating token: Not applicable, a guess is possible.Confidence Score: 5/5Strategizing complete. I will now generate the response based on this analysis.spring annotations

    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_38_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
     envío
    0.44
    env
    0.41
     zav
    0.38
    铺
    0.35
    鍗
    0.35
    descriptors
    0.34
     प्री
    0.33
    Roe
    0.33
    Dil
    0.33
     vehículos
    0.33
    POSITIVE LOGITS
    navbarNav
    0.39
    raquo
    0.38
    拡散方程式
    0.37
    CompoundButton
    0.36
    uasion
    0.36
    ,))
    0.36
     నే
    0.36
    ผู้ตัดสิน
    0.35
     Mercier
    0.35
     lantern
    0.34
    Activations Density 0.000%

    No Known Activations