Neuron Explorer

Qwen3-0.6B

'is' after a long subject phrase

A neuron in layer 18 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the word 'is' when it comes after a long subject, like 'understanding how to create a function in Arduino is vital'. It is a grammar-position detector. Turning it up made the model's writing drift into rambling, unrelated sentences.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    functions of an Arduino offer utility of using subroutines. Understanding how to create a function in Arduino is vital for any programmer out there. With that said, we shall be discussing how

  2. #2 peak

    development. They create financial indicators for text mining. They gather text and analyze the content to produce market indicators. For some data science jobs, knowing about machine learning is necessary.⏎Machine Learning Engineers⏎have master’s degrees, and sometimes PhDs in

  3. #3 peak

    will work as well.⏎Utilizing your BFFs.⏎Having the right friends and someone to talk to and count on is extremely important for your mental health. Seek out groups and activities that will attract new

  4. #4 peak

    katakana and includes the English words “fox” and “face,” would seem to have been borrowed from abroad. But the plant isn’t called fox face anywhere outside Japan; this is purely wasei eigo (made-in-Japan

  5. #5 peak

    by the seemingly complicated written structure. With Music Theory for Dummies, understanding music has never been easier!⏎Note: CD-ROM/DVD and other supplementary materials are not included as part of eBook file.<|endoftext|>is a former municipality in Akershus

  6. #6 peak

    Please schedule an appointment through Connect.⏎Getting a flu vaccine and a COVID-19 vaccine--including a COVID-19 booster shot--at the same time is safe, says Adam R. Stracher, MD, chief medical officer at W

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

turned down a lot
Prompt

Here is something I have been thinking about lately: I have a need to always and always and and and and and.........................................................................................................

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    at the question, as well as all of the answer options before reading a passage. Skimming the question and the answer options gives a student an idea of what to look for in the passage. Perhaps the question concerns the main idea of the passage

  2. #2 peak

    ; a noble man does not love the world or anything in the world. Proverbs 31:8-9 -- a noble man stands up for what is right and shows concern for others. Such are the sayings of King Lemuel

  3. #3 peak

    , guided with PHD Guide⏎HA 6 x 15 minutes<|endoftext|>Have you ever sat down to begin writing and not have a single idea about what to write? We are learning to generate ideas in different ways so that when we sit

  4. #4 peak

    image, created using an online image generator.⏎This wasn’t the only piece of misinformation shared during and immediately after the terrorist attack. However, the majority had one thing in common: a visual element. Rarely were stories shared exclusively in text

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L17:1193 suppresses
  2. L17:2676 excites
  3. L16:1500 suppresses
  4. L17:1796 suppresses
  5. L14:630 excites
  6. L15:2366 suppresses
  7. L15:618 suppresses
  8. L14:2086 suppresses
  9. L17:733 suppresses
  10. L15:1656 excites
  11. L17:2175 excites
  12. L13:602 suppresses
  13. L17:2109 suppresses
  14. L17:503 suppresses
  15. L14:1473 excites
  16. L15:1692 excites
  17. L12:2414 suppresses
  18. L15:1537 excites
  19. L14:1956 suppresses
  20. L16:2143 suppresses
  21. L17:294 excites
  22. L15:1602 suppresses
  23. L17:2879 excites
  24. L12:1271 suppresses

Feeds later layers

  1. L19:2412 suppresses
  2. L19:1045 suppresses
  3. L19:716 suppresses
  4. L19:1549 excites
  5. L20:1726 excites
  6. L19:2798 suppresses
  7. L19:544 excites
  8. L20:1152 suppresses
  9. L19:1488 suppresses
  10. L19:1371 suppresses
  11. L22:2626 excites
  12. L22:181 excites
  13. L20:1561 suppresses
  14. L20:585 excites
  15. L20:1093 suppresses
  16. L19:566 weakly excites
  17. L20:1932 weakly suppresses
  18. L24:422 weakly excites
  19. L19:2103 weakly suppresses
  20. L20:2187 weakly suppresses
  21. L22:1971 weakly suppresses
  22. L20:1771 weakly suppresses
  23. L19:3049 weakly suppresses
  24. L19:2236 weakly suppresses

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

1 values between -8.83 and -8.451 values between -5.42 and -5.051 values between -5.05 and -4.671 values between -4.67 and -4.291 values between -3.91 and -3.534 values between -3.53 and -3.169 values between -3.16 and -2.788 values between -2.78 and -2.4016 values between -2.40 and -2.0225 values between -2.02 and -1.6448 values between -1.64 and -1.2692 values between -1.26 and -0.89192 values between -0.89 and -0.51585 values between -0.51 and -0.131551 values between -0.13 and 0.25581 values between 0.25 and 0.63336 values between 0.63 and 1.01172 values between 1.01 and 1.38101 values between 1.38 and 1.7669 values between 1.76 and 2.1443 values between 2.14 and 2.5230 values between 2.52 and 2.9016 values between 2.90 and 3.287 values between 3.28 and 3.659 values between 3.65 and 4.034 values between 4.03 and 4.411 values between 4.41 and 4.791 values between 4.79 and 5.171 values between 5.17 and 5.541 values between 5.54 and 5.921 values between 5.92 and 6.30offnegativepositiveoften

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.

Promotes

  1. etical
  2. travel
  3. 被
  4. 工作
  5. travels
  6. ipro
  7. 已被
  8. losed
  9. mism
  10. ival

Suppresses

  1. reak
  2. vez
  3. sez
  4. edian
  5. 就想
  6. 不排除
  7. DataContext
  8. }->{
  9. containing
  10. implementing