Neuron Explorer

Qwen3-0.6B

end of sentence before a claim or statistic

A neuron in layer 21 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 period or question mark that ends a sentence introducing a point, fact or claim, where the next sentence is about to explain or support it. When turned up, the model's writing became a bit more repetitive and list-like, and when turned down it produced stray asterisks and repeated symbols.

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

    , saving energy and reducing utility costs.⏎Choosing the Right Replacement Parts⏎When it comes to repairing or upgrading an HVAC system, choosing the correct replacement parts is critical. Quality parts ensure compatibility with the system and extend its lifespan. For example, using

  2. #2 peak

    When it comes to game design, it's mainly about creating a world that imitates their own world and letting them explore it in a similar way. She adds that if you watch young children playing with blocks, the breaking down is

  3. #3 peak

    grain conveyor and then discharged into the grain tank.<|endoftext|>Urine Smells Bad for These Eight Reasons⏎Have you noticed that your urine always has a bad smell? Do you want to know the reasons why urine smells bad? If so, keep

  4. #4 peak

    your self !⏎Conspiracies: The Pyramid and the Sphinx⏎By Steven N. Ng⏎Most of the conspiracy theories revolving around the Sphinx usually involve its age. The Sphinx is generally accepted to have been built along with the Pyramids of G

  5. #5 peak

    a sign of a more serious medical condition.⏎What Causes Loss of Appetite in Seniors?⏎There are a variety of reasons why seniors may have decreased appetites. The non-medical causes of loss of appetite in the elderly include:⏎- Changing

  6. #6 peak

    .⏎Sometimes we allow the connections to be 'forced' on us. For example, if you were to be asked where you were last Wednesday at 15.00 hours, chances are you wouldn't immediately recall. But if you were

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:** ** ** **** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **

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

    “Helping students to increase their vocabulary and to feel comfortable using words in various contexts is paramount,” state Education Commissioner Armando Vilaseca said in a statement. “There is significant

  2. #2 peak

    the Philippines are trying to manage this incident carefully,” Jonah Blank, senior political scientist specializing in Southeast Asia for RAND Corp., a global policy think-tank, tells TIME. ”We’ve seen many Muslim rebel groups arise or take refuge in the southern part

  3. #3 peak

    George Washington University School of Medicine and Health Sciences. âWe hope health care providers will take note and prescribe this approach to patients looking to manage their weight and health.â⏎One of the secrets behind losing weight on a plant-based diet is to

  4. #4 peak

    going to happen really quite gradually for a long time yet.” Professor Atkinson said although automation had the potential to take some jobs it could also provide new employment opportunities. “You’re not going to be using machines to build new mines, people will

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. L15:403 suppresses period inside an abbreviation
  2. L16:432 suppresses period after month or initial
  3. L17:347 excites full stops ending informative sentences
  4. L18:791 suppresses
  5. L19:344 excites periods ending explanations
  6. L4:7 excites
  7. L20:460 suppresses
  8. L9:16 suppresses unclear: periods inside numbers and names
  9. L18:961 suppresses
  10. L11:132 excites period ending an opening topic sentence
  11. L10:612 excites full stop ending boilerplate or a title
  12. L19:711 suppresses
  13. L19:1820 excites
  14. L5:155 suppresses
  15. L17:1772 suppresses
  16. L18:832 suppresses
  17. L20:166 suppresses
  18. L18:2043 excites
  19. L17:839 suppresses
  20. L20:720 suppresses
  21. L18:450 excites
  22. L18:160 suppresses
  23. L17:856 suppresses
  24. L18:888 suppresses

Feeds later layers

  1. L25:40 strongly excites
  2. L27:1801 suppresses
  3. L25:297 suppresses
  4. L27:1496 suppresses
  5. L23:171 suppresses
  6. L26:167 suppresses
  7. L22:2621 suppresses
  8. L22:165 suppresses
  9. L26:1018 suppresses
  10. L24:30 suppresses
  11. L25:533 suppresses first letter m or oct fragments
  12. L23:337 suppresses
  13. L25:334 suppresses
  14. L25:940 excites
  15. L27:637 suppresses
  16. L23:444 suppresses
  17. L27:952 suppresses
  18. L22:328 suppresses
  19. L25:91 suppresses
  20. L24:145 weakly suppresses
  21. L27:713 weakly suppresses
  22. L25:456 weakly suppresses
  23. L26:11 weakly suppresses
  24. L24:12 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 -15.02 and -14.251 values between -13.47 and -12.691 values between -9.58 and -8.802 values between -8.02 and -7.243 values between -7.24 and -6.462 values between -6.46 and -5.689 values between -4.91 and -4.138 values between -4.13 and -3.3520 values between -3.35 and -2.5731 values between -2.57 and -1.7968 values between -1.79 and -1.01399 values between -1.01 and -0.242124 values between -0.24 and 0.54544 values between 0.54 and 1.32278 values between 1.32 and 2.10140 values between 2.10 and 2.88101 values between 2.88 and 3.6648 values between 3.66 and 4.4438 values between 4.44 and 5.2132 values between 5.21 and 5.9926 values between 5.99 and 6.778 values between 6.77 and 7.556 values between 7.55 and 8.333 values between 8.33 and 9.104 values between 9.10 and 9.884 values between 9.88 and 10.662 values between 10.66 and 11.442 values between 11.44 and 12.221 values between 12.22 and 13.002 values between 15.33 and 16.11offnegativepositiveoften

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. ...";⏎
  2. ..."); ⏎
  3. :@{
  4. ."; ⏎
  5. .";⏎
  6. .";
  7. ...');⏎
  8. ...");⏎
  9. ...",
  10. ,...⏎⏎

Suppresses

  1. ⏎
  2. #⏎
  3. ⏎⏎
  4. ###⏎
  5. ⏎⏎⏎
  6. **⏎
  7. ####
  8. </
  9. ⏎