Neuron Explorer

Qwen3-0.6B

space right before a number

A neuron in layer 17 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 spot just before a number is about to appear, such as 'DAYS 2-6', 'over the age of 18' or 'Section 1'. Its strongest negative values also come in number ranges, so the sign is not simple. Turning it up made the model repeat words and drift into lists.

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

    inerary⏎- DAY 1 Arrival, travel to accommodations, welcome presentation⏎- DAYS 2-6 Excavations, catalog artifacts, map excavation sites⏎- DAY

  2. #2 peak

    a system which has been stressed beyond a certain level does not return to its original state when that stress is removed but migrates to a new state.” (p.85) He uses the analogy of stretching and overloading a wire spring (

  3. #3 peak

    4 weeks of focused work, and then reassessing to check that progress has been made in key areas.⏎- Week 1: Initial assessment;⏎- Weeks 2 -5: One English, maths, and science lesson-based project per week

  4. #4 peak

    vote. Also a place where government is by the people and for the people. The United Kingdom is a representative democracy. This means that everyone over the age of 18 can vote in elections, and the people they elect then represent their interests

  5. #5 peak

    given to civilians and less power given to military officials. There are many articles contained within this document, and here are some of the main ones:⏎1) Section 1 – The people shall be sovereign;⏎2) Article

  6. #6 peak

    eleven hour work day was the norm (60 hours per week). Look at this time schedule: Breakfast at 6:15, Commence Work at 6:45, Take an hour for mid-day meal, return to work

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: I have a time that is equal to the value of the value of the value of the value of the value of the value of the value of the value of the value of the value of the value of the value of the value of 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

    with 500 being the highest score. Grade-level/subject-level scores are given in terms of five achievement levels, with 1 being the lowest and 5 the highest. The writing test is scored on a scale of 1 to

  2. #2 peak

    .⏎The values are stored in sequential order in an array and we can access these elements or values by using the index. The index starts from 0 and ends at length of the array - 1.⏎i.e. the index of the first

  3. #3 peak

    , Jupiter, Saturn.⏎The 12 sectors represented the zodiac houses. Every sector was divided into a number of square, from 1 for Moon to 7 for Saturn.⏎The

  4. #4 peak

    USDA and FDA tests have produced hard evidence of widespread presence of pesticide residues on conventional crops. The most recent round of tests show that as late as 2010, 68 percent of food samples had detectable pesticide residues. EW

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. L15:2294 suppresses
  2. L11:1266 suppresses
  3. L16:372 excites
  4. L15:664 weakly suppresses
  5. L8:350 weakly suppresses blank space before a year
  6. L3:2372 weakly suppresses
  7. L11:1004 weakly excites
  8. L14:2281 weakly excites
  9. L11:3064 weakly excites
  10. L16:3031 weakly excites
  11. L1:1759 weakly excites
  12. L16:2963 weakly excites
  13. L11:1532 weakly suppresses
  14. L8:1631 weakly excites
  15. L10:969 weakly suppresses
  16. L16:276 weakly excites
  17. L10:1623 weakly suppresses
  18. L2:2922 weakly excites
  19. L3:1564 weakly suppresses
  20. L8:2510 weakly excites
  21. L13:1083 weakly excites
  22. L10:2693 weakly excites
  23. L10:1818 weakly suppresses
  24. L8:2591 weakly excites

Feeds later layers

  1. L20:1994 excites
  2. L22:717 suppresses
  3. L18:2822 suppresses
  4. L21:2426 excites
  5. L19:412 weakly excites
  6. L19:2012 weakly suppresses
  7. L20:670 weakly excites
  8. L19:1736 weakly excites
  9. L27:2903 weakly excites
  10. L19:2735 weakly excites
  11. L18:1899 weakly excites
  12. L18:2573 weakly suppresses
  13. L25:1910 weakly excites
  14. L21:1120 weakly excites
  15. L22:1553 weakly excites
  16. L19:1134 weakly suppresses
  17. L20:1768 weakly excites
  18. L25:2579 weakly excites
  19. L19:1285 weakly suppresses
  20. L18:2547 weakly excites
  21. L20:1682 weakly suppresses
  22. L27:1004 weakly excites
  23. L21:2354 weakly excites
  24. L18:2491 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 -3.34 and -3.211 values between -2.95 and -2.811 values between -2.29 and -2.162 values between -1.76 and -1.631 values between -1.63 and -1.502 values between -1.50 and -1.371 values between -1.37 and -1.241 values between -1.24 and -1.112 values between -1.11 and -0.971 values between -0.97 and -0.844 values between -0.84 and -0.719 values between -0.71 and -0.5843 values between -0.58 and -0.45136 values between -0.45 and -0.32360 values between -0.32 and -0.191012 values between -0.19 and -0.051520 values between -0.05 and 0.08576 values between 0.08 and 0.21161 values between 0.21 and 0.3445 values between 0.34 and 0.4717 values between 0.47 and 0.608 values between 0.60 and 0.742 values between 0.74 and 0.871 values between 1.52 and 1.661 values between 1.79 and 1.92offnegativepositiveoften

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. hon
  2. next
  3. NEXT
  4. 后续
  5. 起步
  6. _NEXT
  7. 下一代
  8. luck
  9. _Start
  10. 后面的

Suppresses

  1. 万余元
  2. AreaView
  3. ague
  4. endo
  5. ertext
  6. ивают
  7. nage
  8. (completion
  9. _deck
  10. vably