Neuron Explorer

Qwen3-0.6B

comma before a participle phrase

A neuron in layer 20 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 a comma that comes after a clause like "and" or "then", just before the sentence continues with a phrase such as "finding his enemy dead". Turning it up made the model stutter on the word "using", while turning it down gave ordinary text.

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

    by Julius Caesar in Greece, fled to Egypt seeking solace but was immediately murdered by agents of Ptolemy XIII. Caesar arrived in Alexandria soon after and, finding his enemy dead, decided to restore order in Egypt.⏎During the preceding century

  2. #2 peak

    amusement.⏎There was a half-decayed log lying by the roadside, and Grant sat down on it, calmly pulled a cigar from his cigar case, and, requesting one of his staff who stood nearby to furnish him a sheet of paper,

  3. #3 peak

    Hebrides that Columba turned his face when he accepted the terrible penance of Molaise. He bade farewell to his relatives, and, with a few monks who insisted on accompany him whithersoever he might

  4. #4 peak

    put his scalp on display for money.⏎In 1900, Thompson sent the scalp back to the Nebraska doctor who tried to reaffix it, who, in turn, donated it to the public library. The scalp remained on display there

  5. #5 peak

    posited answers to this question, whether directly - in the case of Plato and Plotinius - or through story - in the case of Homer.⏎Plotinus, having been a proponent of Platonism, naturally concurs with Plato,

  6. #6 peak

    built except the roof," Thomas replied. So Gundaphorus sent him gold and silver to roof the palace, and Thomas, thanking<|endoftext|>The Internet and copyright⏎Choose a topic:⏎Internet technology is developing faster

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: the design of a software system which is not only created to create but also to create create... create and create create... create create. It is so created, so create and create. It is also created to create create... create create.

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

    ,0<|endoftext|>Guidelines for Restoring Cultural Landscapes⏎Identifying, retaining and preserving

  2. #2 peak

    1,700.<|endoftext|>Risk assessment for construction⏎The Construction (Design and Management) Regulations (CDM

  3. #3 peak

    A white turkey is a domesticated turkey that has been bred to have white feathers. White turkeys are not albino; they have black eyes and dark skin. The Leycester sub-group of the Bronze turkey is the only true genetic white

  4. #4 peak

    others only shame and hardship.⏎This I can say for myself:⏎that for awhile I was the Heodeninga<|endoftext|>This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.⏎As

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. L19:94 suppresses
  2. L19:2683 suppresses
  3. L19:2709 excites
  4. L18:2434 excites
  5. L17:3068 excites comma after a sentence subject
  6. L19:788 weakly excites
  7. L15:835 weakly suppresses
  8. L16:1808 weakly excites
  9. L11:2200 weakly excites
  10. L2:2775 weakly excites
  11. L19:2707 weakly excites
  12. L17:2041 weakly suppresses
  13. L0:509 weakly suppresses
  14. L4:1079 weakly suppresses
  15. L19:1819 weakly excites
  16. L19:119 weakly suppresses
  17. L17:1209 weakly suppresses
  18. L18:1926 weakly suppresses
  19. L12:1445 weakly suppresses
  20. L11:1045 weakly suppresses
  21. L9:1137 weakly excites
  22. L3:2958 weakly excites
  23. L17:1839 weakly suppresses
  24. L3:1908 weakly suppresses

Feeds later layers

  1. L24:140 strongly excites
  2. L22:1425 strongly excites
  3. L25:376 strongly excites
  4. L26:252 strongly excites
  5. L26:686 strongly excites
  6. L27:410 strongly excites
  7. L27:1237 strongly excites
  8. L22:609 excites
  9. L27:1890 excites
  10. L25:1596 excites
  11. L27:303 excites
  12. L27:1148 excites
  13. L25:218 excites
  14. L26:708 excites
  15. L23:424 excites
  16. L25:620 excites
  17. L27:677 excites
  18. L26:1304 excites
  19. L27:1545 excites
  20. L21:1630 excites
  21. L21:615 excites
  22. L27:1852 excites
  23. L26:733 excites
  24. L26:852 excites

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 -7.15 and -6.661 values between -6.16 and -5.663 values between -5.17 and -4.673 values between -4.67 and -4.1810 values between -4.18 and -3.688 values between -3.68 and -3.188 values between -3.18 and -2.6910 values between -2.69 and -2.1917 values between -2.19 and -1.7021 values between -1.70 and -1.2057 values between -1.20 and -0.70606 values between -0.70 and -0.212516 values between -0.21 and 0.29544 values between 0.29 and 0.7840 values between 0.78 and 1.289 values between 1.28 and 1.7810 values between 1.78 and 2.2711 values between 2.27 and 2.774 values between 2.77 and 3.266 values between 3.26 and 3.764 values between 3.76 and 4.262 values between 4.26 and 4.752 values between 4.75 and 5.253 values between 5.25 and 5.741 values between 5.74 and 6.242 values between 6.24 and 6.742 values between 7.23 and 7.731 values between 7.73 and 8.221 values between 8.22 and 8.721 values between 9.71 and 10.211 values between 10.70 and 11.202 values between 11.70 and 12.191 values between 12.19 and 12.69offnegativepositiveoften

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. using
  2. using
  3. -using
  4. relying
  5. applying
  6. considering
  7. _using
  8. taking
  9. recognizing
  10. preferring

Suppresses

  1. nữa
  2. ocumented
  3. 其
  4. duce
  5. 那就是
  6. create
  7. Describe
  8. cả
  9. CloseOperation
  10. Creates