Neuron Explorer

Qwen3-0.6B

comma after a sentence subject

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 right after a noun phrase, where the sentence is about to add an extra detail, like 'this site[,] using published books' or 'children[,] aged 14'. Turned up, the model wrote text with no clear shared theme, so steering does not support the pattern.

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

    |⏎Definitions, pronunciations, and word histories. Usage of modern and historical words traced through 2.5 million quotations.⏎Enter any word and this site, using published reference books, provides⏎rhymes, synonyms, antonyms, definitions

  2. #2 peak

    time of a maltreatment investigation associated with the likelihood of remaining in foster care? NSCAW is a national longitudinal study of 5,501 children, aged 14 years or younger, who had contact with the child welfare system

  3. #3 peak

    Department.⏎The two-dimensional material used is called transition-metal dichalcogenides (TMDs). Not only is it small, it also allows for faster electron transfer — aka, faster charging and discharging —

  4. #4 peak

    ’ll know all about the need for managing the formation of hydrates during the pumping and processing of natural gas, but there’s now a new means for preventing such, and it’s a lot more friendly to the environment and company budgets.⏎That’s

  5. #5 peak

    class and outside of it. However, it takes time, certain skills, and a lot of work.⏎Detailed plan on how to be more

  6. #6 peak

    Testing (PhD in Physical Science)⏎A product tester is someone who tests a product before it gets released to the general public. Testing specialists examine and analyze products, identifying potential flaws. Product testers might work for a company

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: how to help the ****************************************************************************************

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

    something to color with⏎The directions are appropriate for an older child or adult to⏎assemble. Don't worry too much if younger children glue the pieces a bit⏎off kilter -- part of the fun of watching them grow is seeing their⏎

  2. #2 peak

    its component parts was unthinkable, impossible -- a conception as striking⏎as it was new to political theory.'"⏎"The Archbishop of York said recently:

  3. #3 peak

    900 razorback sucker and 438 bonytail from these releases, but only 186 and 76 of these contacts⏎<|endoftext|>Blue Beauty-

  4. #4 peak

    0)⏎See Corbett (2000:173) for languages where mass nouns take plural rather⏎than singular morphology (and a few languages where there is a split)⏎Swedish

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. L18:299 suppresses
  2. L19:35 excites
  3. L17:310 suppresses
  4. L16:1258 suppresses
  5. L17:1567 excites
  6. L18:237 suppresses
  7. L16:688 suppresses
  8. L18:933 suppresses
  9. L19:1363 suppresses
  10. L18:1174 suppresses
  11. L17:2017 suppresses
  12. L18:169 excites periods ending informational sentences
  13. L19:2368 suppresses
  14. L7:2384 excites
  15. L15:933 excites
  16. L8:216 excites
  17. L18:116 excites
  18. L19:2818 weakly excites
  19. L17:2093 weakly excites
  20. L17:933 weakly excites
  21. L16:2227 weakly suppresses
  22. L13:1108 weakly suppresses
  23. L18:257 weakly suppresses
  24. L19:726 weakly suppresses comma inside large numbers

Feeds later layers

  1. L21:348 strongly suppresses mid-sentence line breaks
  2. L25:428 weakly suppresses
  3. L26:592 weakly excites
  4. L27:146 weakly excites
  5. L21:594 weakly suppresses
  6. L25:469 weakly suppresses
  7. L24:48 weakly excites
  8. L25:210 weakly suppresses
  9. L21:200 weakly excites
  10. L21:875 weakly suppresses
  11. L27:100 weakly excites
  12. L26:357 weakly suppresses
  13. L27:938 weakly suppresses
  14. L21:360 weakly suppresses 'such' introducing examples
  15. L27:451 weakly excites
  16. L25:334 weakly excites
  17. L25:2859 weakly excites
  18. L26:826 weakly suppresses
  19. L26:372 weakly excites
  20. L21:581 weakly excites
  21. L25:197 weakly suppresses
  22. L22:341 weakly suppresses
  23. L21:1854 weakly suppresses
  24. L25:1094 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 -18.69 and -18.031 values between -18.03 and -17.371 values between -10.81 and -10.151 values between -8.84 and -8.181 values between -7.53 and -6.871 values between -6.87 and -6.212 values between -6.21 and -5.562 values between -5.56 and -4.906 values between -4.90 and -4.248 values between -4.24 and -3.5910 values between -3.59 and -2.939 values between -2.93 and -2.2723 values between -2.27 and -1.6249 values between -1.62 and -0.96554 values between -0.96 and -0.302761 values between -0.30 and 0.35259 values between 0.35 and 1.01109 values between 1.01 and 1.6756 values between 1.67 and 2.3216 values between 2.32 and 2.9817 values between 2.98 and 3.648 values between 3.64 and 4.298 values between 4.29 and 4.951 values between 5.61 and 6.262 values between 6.26 and 6.922 values between 6.92 and 7.57offnegativepositiveoften

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. -on
  2. .
  3. On
  4. typical
  5. small
  6. In
  7. total
  8. recent
  9. _
  10. み

Suppresses

  1. |RF
  2. owie
  3. dings
  4. ableView
  5. orque
  6. Swinger
  7. 斩获
  8. ).__
  9. ews
  10. зе