Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsExportsSlackBlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Qwen3-4B
    3. 0-TRANSCODER-HP
    4. 2
    Prev
    Next
    INDEX
    Explanations

    was and quotation marks

    np_max-act · gemini-2.0-flash

    instances of past‑tense copular/auxiliary constructions, especially in “what … was” frames, with some sensitivity to adjacent definite articles.

    oai_token-act-pair · gpt-5Triggered by @liuliujiujiu05
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    mwhanna/qwen3-4b-transcoders/layer_0.safetensors
    Prompts (Dashboard)
    16,384 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    163,840
    Data Type
    float32
    Hook Name
    blocks.0.mlp.hook_in
    Architecture
    transcoder
    Context Size
    8,192
    Dataset
    monology/pile-uncopyrighted
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    lat
    -0.28
    eness
    -0.28
    uv
    -0.28
    æĺ¯éŁ©åĽ½
    -0.27
    è¢ĭåŃIJ
    -0.26
    èĥ½æĬĬ
    -0.24
    æ·¡åĮĸ
    -0.24
    osos
    -0.24
    æķ·
    -0.24
     organs
    -0.23
    POSITIVE LOGITS
     racked
    0.30
    klä
    0.28
    樵
    0.27
    uter
    0.27
    beat
    0.26
    æĹ¶éļĶ
    0.26
    ä¸Ģéģį
    0.25
    ç»°
    0.25
     risking
    0.25
    anager
    0.25
    Activations Density 8.270%

    No Known Activations