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

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsExportsSlackBlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Joseph Bloom · Initial Gemma-2B-IT SAEs
    3. Gemma-2B-IT
    4. Residual Stream
    5. 12-RES-JB
    6. 2917
    Prev
    Next
    INDEX
    Explanations

    discussions related to political and legislative struggles, specifically regarding changes in laws and policies

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot

    words related to complex or technical terminology, especially involving policy, legislation, and administrative language.

    oai_token-act-pair · claude-3-5-haiku-20241022Triggered by @eekay

    word fragments that are broken by hyphens, particularly syllable breaks in words.

    oai_token-act-pair · claude-3-7-sonnet-20250219Triggered by @eekay

    Subword tokenization patterns where words are segmented into smaller linguistic units by a tokenizer, visible as partial syllables and morphemes across diverse text contexts.

    eleuther_acts_top20 · claude-3-7-sonnet-20250219Triggered by @eekay
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    jbloom/Gemma-2b-IT-Residual-Stream-SAEs/gemma_2b_it_blocks.12.hook_resid_post_16384
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    HuggingFaceFW/fineweb
    Features
    16,384
    Data Type
    float32
    Hook Name
    blocks.12.hook_resid_post
    Hook Layer
    12
    Architecture
    standard
    Context Size
    1,024
    Dataset
    Skylion007/openwebtext
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    876
    +0.11
    0.3%
    1499
    +0.10
    0.3%
    674
    +0.09
    0.3%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    392
    +0.11
    0.01
    1401
    +0.10
    0.02
    1032
    +0.09
    0.02
    Negative Logits
    Czym
    -0.69
    Eksteraj
    -0.67
    Ekster
    -0.64
    Răsp
    -0.64
    Literat
    -0.64
    Dicas
    -0.61
    Și
    -0.61
    Weiterlesen
    -0.60
    Kleur
    -0.60
    Embal
    -0.60
    POSITIVE LOGITS
     dises
    1.31
    <bos>
    1.31
     dispen
    1.27
     erec
    1.26
     effe
    1.24
     seiz
    1.20
     nece
    1.19
     haer
    1.19
     haup
    1.19
     tamen
    1.19
    Activations Density 0.054%

    No Known Activations