Neuron Explorer

Qwen3-0.6B

spot where a number is due

A neuron in layer 25 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 switches on just before a number is about to appear, such as a percentage, a price, a calorie count or a year-like quantity. When it was turned up, the model's text drifted into odd repeated fragments, so the steering tells us little about it.

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

    a whole. Unfortunately, one hundred years later, women still make up a majority of the world’s poor.⏎Women’s wages in the U.S. are only 76% of men’s, but the disparity of long-term earnings between men

  2. #2 peak

    12th Social Studies & History⏎Anticipating the State of the Union, January 2010⏎Designed to precede President Obama’s 2

  3. #3 peak

    helps maintain a healthy digestive system and promotes regular bowel movements. Eating a 100-gram serving of tamales introduces 6.28

  4. #4 peak

    and orchards are the best choices for optimal nutrients, energy and vitality. Fresh fruits, vegetables, nuts, seeds, whole grains and legumes should ideally constitute 60-70% of your diet. Although processed food has an undeniable

  5. #5 peak

    greater level of experience, and less time spent raising children. When President John F. Kennedy signed the Equal Pay Act into law, women earned, on average, 60 cents for every dollar earned by men. In the forty-eight years that

  6. #6 peak

    If you're using a calorie counter, remember that Fat, Carbs, and Protein calories are just close estimations based on the Atwater factors:⏎Fat: 9 cal/g Carb: 4 cal/g Protein: 4 cal/g⏎

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: Iasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasasas

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

    blood vessels (pyelonephritis, glomerulonephritis, amyloidosis, heart failure, renal vascular anomalies).⏎Reduction of

  2. #2 peak

    care provider checks for the disorder by using techniques called amniocentesis (pronounced am-nee-oh-sen-TEE-sis) or chorionic (pronounced kawr-ee-ON-ik) villus sampling

  3. #3 peak

    alopathy-PML. Updated January 8, 2016. Accessed September 26, 2016.⏎Progressive multifocal leukoencephalopathy (PML). AETC National Resource Center website

  4. #4 peak

    The best way to get rid of H. pylori is to use triple-therapy for 10 to 14 days. Usually antibiotics such as clarithromycin, amoxicillin and

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. L21:339 excites spot just before a year
  2. L19:581 suppresses blank slot before a number
  3. L23:77 excites the space right before a number
  4. L18:377 excites number right after 'about' or 'less than'
  5. L22:372 excites empty spot before a number or "3D"
  6. L22:824 excites blank before an expected number
  7. L4:53 excites blank slot before a number
  8. L17:170 excites
  9. L20:436 suppresses
  10. L8:111 suppresses the space before a year in dates
  11. L17:285 weakly suppresses
  12. L21:613 weakly suppresses
  13. L22:216 weakly excites
  14. L18:93 weakly suppresses
  15. L16:314 weakly excites
  16. L13:923 weakly suppresses word right before a number or year
  17. L0:2567 weakly suppresses
  18. L4:1635 weakly suppresses 'light' as an adjective (pale, not heavy)
  19. L14:616 weakly suppresses
  20. L13:1372 weakly excites number before an age or week count
  21. L18:91 weakly suppresses
  22. L14:196 weakly excites
  23. L1:2159 weakly suppresses
  24. L18:203 weakly suppresses

Feeds later layers

  1. L26:103 excites blank space before a number
  2. L27:170 excites unclear: spaces in mixed-language text
  3. L27:283 excites the spot right before a number
  4. L26:15 suppresses
  5. L27:220 excites end-of-text token between documents
  6. L26:121 excites
  7. L27:48 weakly suppresses
  8. L26:216 weakly excites blank slot before a year or number
  9. L26:423 weakly suppresses
  10. L26:145 weakly excites
  11. L27:183 weakly suppresses
  12. L27:2249 weakly suppresses
  13. L26:97 weakly suppresses
  14. L27:265 weakly excites
  15. L26:172 weakly excites
  16. L26:769 weakly suppresses
  17. L26:17 weakly suppresses
  18. L27:238 weakly excites
  19. L26:224 weakly suppresses
  20. L27:1025 weakly suppresses
  21. L26:168 weakly excites
  22. L27:2604 weakly suppresses
  23. L27:317 weakly excites
  24. L27:276 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.

10 values between -13.02 and -7.9763 values between -7.97 and -2.921301 values between -2.92 and 2.141567 values between 2.14 and 7.19675 values between 7.19 and 12.25203 values between 12.25 and 17.3041 values between 17.30 and 22.357 values between 22.35 and 27.415 values between 27.41 and 32.463 values between 32.46 and 37.511 values between 37.51 and 42.571 values between 42.57 and 47.621 values between 47.62 and 52.682 values between 52.68 and 57.731 values between 62.78 and 67.841 values between 67.84 and 72.891 values between 77.94 and 83.001 values between 83.00 and 88.051 values between 88.05 and 93.102 values between 93.10 and 98.162 values between 103.21 and 108.272 values between 108.27 and 113.325 values between 113.32 and 118.373 values between 118.37 and 123.432 values between 123.43 and 128.481 values between 128.48 and 133.531 values between 133.53 and 138.592 values between 143.64 and 148.702 values between 168.91 and 173.961 values between 184.07 and 189.13offnegativepositiveoften

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. iced
  2. ionic
  3. ersions
  4. icens
  5. unted
  6. erts
  7. erti
  8. umber
  9. ices
  10. uther

Suppresses

  1. NIL
  2. yl
  3. ⏎→→⏎
  4. rag
  5. miss
  6. eid
  7. ={()=>
  8. ⏎→⏎
  9. mise
  10. corev