Neuron Explorer

Qwen3-0.6B

digits inside numbers and statistics

A neuron in layer 3 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron reacts to a single digit or decimal point in the middle of a number, such as the "0" in 1703 or the "9" in 1999, mostly in years, percentages and statistics. Turning it up did not produce a clear theme in the model's writing.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    1655, 1655, 1734, 1650, 1733, 1703 and 1707 are included under the topic Early Cox History in

  2. #2 peak

    these tests have excellent detection rates (~99%) with very low false-positive rates (~0.2%). In other words, these tests are about 99.0% sensitive and 99.8% specific.⏎With performance like that

  3. #3 peak

    a priority of ideological harmony, multidimensionality, association and synthesis of previous rationalities for a harmonious civilization and a harmonious person.⏎2. Instead of the multivolume expressions accessible solely to the old age “spiritual aristocrats”

  4. #4 peak

    008, at $94.81 per barrel. That is not far below the average high of 2008, at $99.67. In between was a crash in oil prices — and most commodities

  5. #5 peak

    Commission data, at the traditional poverty line, the India-wide absolute number of poor was 323 million in 1983, 320 million in 1993-94 and 302 million

  6. #6 peak

    forces cut off U.S. access to hevea--and during the 1970s.⏎Guayule grows much faster than hevea, but the latter still has certain advantages.⏎Hevea is less prone to cracking

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: In the name of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of

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

    width. Hence the text:⏎<P> This is another paragraph, with a sentence that is split over several lines.Will get formatted as:⏎This is another paragraph, with a sentence that is split over several lines.<|endoftext|>Insufficient vocabulary, comprehension

  2. #2 peak

    th century by the King Rut Trak Varman. A French historian, Mi Bri No said that the temple is made of laterite stone. Phnom Da, Takeo is not enclosed by any wall and faces northwards.⏎There are

  3. #3 peak

    a voluntary standard and partly encompasses non-renewable technologies like nuclear and “clean coal.”4⏎And the state recently repealed “Energizing Indiana,”5⏎its energy efficiency program that supported nearly 1,500 jobs and saved

  4. #4 peak

    Accusative Case Accusative Case = Direct Object The direct object receives the action of the verb; it is whats being verbed. I buy a CD. Who or what is being bought?⏎Word changes in English In English, some words

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. L0:39 weakly excites 'the' right after 'On' in 'On the other hand'
  2. L2:2328 weakly suppresses
  3. L0:1535 weakly suppresses
  4. L2:2696 weakly excites
  5. L1:59 weakly excites 'the' after a verb or preposition
  6. L1:2340 weakly suppresses
  7. L0:2407 weakly suppresses
  8. L2:3046 weakly suppresses
  9. L1:1100 weakly suppresses
  10. L0:545 weakly suppresses
  11. L0:1811 weakly suppresses
  12. L1:1662 weakly excites
  13. L2:2834 weakly excites
  14. L1:17 weakly suppresses
  15. L2:2747 weakly suppresses
  16. L2:1036 weakly excites
  17. L0:2806 weakly suppresses
  18. L2:1409 weakly suppresses
  19. L0:235 weakly suppresses
  20. L1:2300 weakly excites
  21. L2:1279 weakly excites
  22. L1:316 weakly suppresses
  23. L1:81 weakly excites
  24. L0:1034 weakly excites

Feeds later layers

  1. L4:155 strongly suppresses
  2. L8:73 strongly suppresses
  3. L4:559 strongly excites
  4. L5:119 strongly suppresses
  5. L5:44 strongly suppresses
  6. L6:169 suppresses
  7. L10:350 suppresses
  8. L5:42 suppresses
  9. L7:85 excites
  10. L6:879 excites
  11. L8:82 excites
  12. L8:484 excites
  13. L5:88 suppresses
  14. L7:106 suppresses
  15. L15:528 suppresses
  16. L7:27 suppresses
  17. L5:107 suppresses
  18. L4:5 suppresses
  19. L12:243 suppresses
  20. L5:1 suppresses
  21. L6:258 suppresses
  22. L6:212 suppresses
  23. L8:208 suppresses
  24. L10:35 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 -1.24 and -1.151 values between -1.15 and -1.051 values between -1.05 and -0.962 values between -0.96 and -0.873 values between -0.87 and -0.771 values between -0.77 and -0.688 values between -0.68 and -0.5813 values between -0.58 and -0.4913 values between -0.49 and -0.3937 values between -0.39 and -0.30107 values between -0.30 and -0.20307 values between -0.20 and -0.11975 values between -0.11 and -0.011497 values between -0.01 and 0.08480 values between 0.08 and 0.18220 values between 0.18 and 0.27109 values between 0.27 and 0.3747 values between 0.37 and 0.4629 values between 0.46 and 0.5520 values between 0.55 and 0.6514 values between 0.65 and 0.747 values between 0.74 and 0.844 values between 0.84 and 0.931 values between 0.93 and 1.033 values between 1.03 and 1.122 values between 1.12 and 1.222 values between 1.22 and 1.312 values between 1.31 and 1.411 values between 1.41 and 1.501 values between 2.45 and 2.54offnegativepositiveoften

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. atures
  2. пит
  3. тки
  4. igor
  5. ered
  6. pering
  7. atorio
  8. vis
  9. headed
  10. 骖

Suppresses

  1. ��
  2. 他们
  3. 他们是
  4. 逞
  5. 他们会
  6. 会觉得
  7. 暧
  8. favourable
  9. 让他们
  10. denominator