Neuron Explorer

Qwen3-0.6B

"by" after a past participle

A neuron in layer 16 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the word "by" when it follows a past-tense or past-participle verb, as in "circled by," "ruled by" or "hounded by." It is weakly active there and strongly negative on other uses of "by," such as "reduces emissions by" or a sentence-opening "By Saturday." Turning it up did not give a clear theme.

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

    Tefnut was depicted in the form of a woman who wears on her head the solar disk circled by two cobras. She holds in her hands the sceptre and ankh.

  2. #2 peak

    become a disese with no cure in Ethiopia and the recurring famine shows the same story, same pictures since mid 1970. Ethiopia is rulled by Live Aid funded rebels who came to power robing the Aid money and they hold

  3. #3 peak

    ahead of the Wehrmacht’s drive west. Hounded by the Gestapo in neutral Lisbon, Archduke Otto and his family came to

  4. #4 peak

    strongly reflects the view advocated by Neil Postman (1979) of creating a “thermostatic

  5. #5 peak

    of the woods ambles along the gap between two parallel rows of planted trees. As it reaches a junction, it freezes, momentarily surprised, but then intrigued, by a dim red light appearing from the base of an adjacent tree. Pausing,

  6. #6 peak

    in San Francisco where we see Robert Childan, owner of an antique shop, struggle to befriend prominent Japanese individuals as well as advance in a society now ordered by Japanese sensibilities. Despite his best efforts, he never quite succeeds. He is

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: trying to creating a creating creating creating creating creating... creating... making creating... creating making creating... creating, creating making creating... creating, by making taking taking, by making taking taking, by making taking... by the evening of the evening,

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

    to say Communized in Urdu as possible so you could understand its correct English to Urdu translation. We encourage everyone to contribute in adding more meanings to MeaningIn Dictionary by adding English to Urdu translations, Urdu to Roman Urdu transliterations and Urdu to English

  2. #2 peak

    LMOP is a voluntary assistance and partnership program that reduces greenhouse gas emissions by supporting landfill gas energy project development.⏎Methane, a primary component of landfill

  3. #3 peak

    8-page paper in Word format. Apply APA standards to citation of sources. Use the following file naming convention: LastnameFirstInitial_M5_A1.doc.⏎By Saturday, August 16, 2014, deliver

  4. #4 peak

    hand in hand.⏎In fact, historians believe that yogurt played a pivotal role in the development of human societies. This is due to two main reasons. First, by allowing milk to be stored for longer, more people were able to enjoy it thanks

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. L3:59 excites
  2. L15:2606 suppresses
  3. L6:1826 excites
  4. L15:39 weakly suppresses
  5. L4:1394 weakly excites
  6. L2:1855 weakly excites
  7. L4:591 weakly suppresses
  8. L6:804 weakly suppresses opening bracket before an acronym
  9. L4:1319 weakly excites the 'B' in vitamin B
  10. L6:185 weakly suppresses
  11. L2:75 weakly suppresses
  12. L3:421 weakly suppresses
  13. L1:2340 weakly suppresses
  14. L5:75 weakly excites
  15. L11:438 weakly excites
  16. L1:1464 weakly excites
  17. L14:95 weakly excites
  18. L1:2801 weakly excites
  19. L2:1918 weakly excites
  20. L6:569 weakly suppresses
  21. L7:605 weakly suppresses
  22. L3:2934 weakly excites
  23. L1:1786 weakly suppresses
  24. L7:19 weakly suppresses

Feeds later layers

  1. L17:1997 suppresses
  2. L17:187 suppresses
  3. L19:617 suppresses
  4. L17:517 suppresses "but" in "not only X, but" phrases
  5. L18:242 suppresses
  6. L26:1395 suppresses
  7. L17:1592 suppresses
  8. L17:1721 suppresses 'by' introducing a method or means
  9. L21:1441 suppresses
  10. L19:385 excites
  11. L19:652 suppresses
  12. L20:762 suppresses
  13. L25:577 excites
  14. L26:1077 weakly suppresses
  15. L19:1608 weakly suppresses
  16. L20:241 weakly suppresses
  17. L27:1067 weakly excites
  18. L18:2140 weakly suppresses
  19. L18:226 weakly suppresses
  20. L19:271 weakly suppresses
  21. L19:1769 weakly suppresses
  22. L18:452 weakly suppresses
  23. L26:945 weakly suppresses
  24. L20:457 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.

2 values between -16.83 and -16.372 values between -16.37 and -15.921 values between -10.91 and -10.462 values between -2.73 and -2.272 values between -2.27 and -1.823 values between -1.36 and -0.9110 values between -0.91 and -0.451087 values between -0.45 and 0.002489 values between 0.00 and 0.46284 values between 0.46 and 0.9126 values between 0.91 and 1.37offnegativepositiveoften

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. elsen
  2. narrower
  3. fronts
  4. ège
  5. encoded
  6. yet
  7. можно
  8. cá
  9. ouched
  10. ToBounds

Suppresses

  1. ambi
  2. squ
  3. ��
  4. 学习
  5. rooms
  6. atk
  7. sy
  8. ico
  9. using
  10. ombat