Neuron Explorer

Qwen3-0.6B

dash in year or number ranges

A neuron in layer 7 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

Fires on the dash that joins two years or numbers, like 2011-2012 or 15-19. Turning it up did not produce any clear theme in the writing.

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

    with the Apiary Inspectors of America and The Bee Informed Partnership.⏎The losses were 42% higher than during the 2011-2012 winter, when 21.9% of colonies died

  2. #2 peak

    -- and an early one. While the article admits that scientists and doctors struggle to predict flu seasons each year, general consensus indicates that the 2017 - 2

  3. #3 peak

    academic needs, as well as any social or emotional concerns they’ve noticed.⏎The high school has been implementing the BARR program since the 2017-2018 school year and has seen significant improvements in course credits earned,

  4. #4 peak

    2017-18 Extension Wheat Production Guide, Georgia farmers planted 160,000 acres in the 2016-2017 season, which marked the third year the state’s wheat acre

  5. #5 peak

    DOE, the Department of Defense and NASA. The company hopes to raise another $2 million by next year, $35 million in 2015-17, and $200 million for its pilot plant stage.⏎The

  6. #6 peak

    , that parents should personally educate their kids. The Scheme of Inclusive Education for Disabled at Secondary Stage (IEDSS) was launched throughout 2009-10 and replaces the earlier scheme of Integrated Education for Disabled Kids (IEDC

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: this is the time of year, and I have to do this. It's important. This is a problem. So. Let's go. It's a good day. Let's not do it. It's the right choice. 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

    vomiting (p < 0.001, OR = 1.969, 95% CI = 1.618-2.398), while fewer manifested fever (p = 0.

  2. #2 peak

    ks (p = 0.001, OR = 1.348, 95% CI = 1.124-1.618) were affected with norovirus when compared with non

  3. #3 peak

    .73–0.97, n = 25) and 0.72 (95% CI = 0.52–0.86; n = 27), respectively. Results indicated that

  4. #4 peak

    (95% CI, 1.4-3.0) for alcohol, 4.2 (95% CI, 3.1-5.8) for marijuana, 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. L4:46 suppresses
  2. L3:101 suppresses
  3. L6:1868 excites
  4. L5:291 excites hyphens inside codes and page ranges
  5. L1:404 weakly suppresses
  6. L1:320 weakly excites
  7. L3:135 weakly suppresses
  8. L6:831 weakly excites
  9. L4:799 weakly suppresses
  10. L2:18 weakly suppresses
  11. L6:2170 weakly excites
  12. L4:25 weakly excites
  13. L6:3009 weakly excites
  14. L2:66 weakly suppresses
  15. L5:1817 weakly suppresses
  16. L3:807 weakly suppresses comma inside dates and years
  17. L0:1911 weakly suppresses
  18. L1:2234 weakly suppresses
  19. L5:2251 weakly suppresses
  20. L6:1235 weakly suppresses
  21. L6:900 weakly suppresses
  22. L6:3050 weakly suppresses
  23. L4:1170 weakly suppresses
  24. L6:2159 weakly suppresses

Feeds later layers

  1. L12:1851 excites hyphen inside dates
  2. L9:1820 excites
  3. L17:364 excites hyphen joining two words
  4. L14:2202 excites
  5. L19:804 excites
  6. L9:589 excites
  7. L15:2247 weakly excites
  8. L15:1042 weakly excites
  9. L15:1549 weakly excites
  10. L9:1031 weakly suppresses
  11. L22:1331 weakly excites
  12. L19:2462 weakly excites
  13. L16:2498 weakly excites
  14. L16:2580 weakly excites 'and' linking food or health nouns
  15. L9:392 weakly excites "and" inside organisation names
  16. L11:2394 weakly excites
  17. L24:201 weakly suppresses
  18. L19:1475 weakly excites
  19. L10:1367 weakly excites
  20. L20:1886 weakly excites
  21. L8:1913 weakly excites
  22. L21:1624 weakly excites
  23. L16:1246 weakly excites
  24. L17:2613 weakly excites

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 -1.45 and -1.321 values between -1.32 and -1.201 values between -1.20 and -1.071 values between -0.94 and -0.823 values between -0.82 and -0.694 values between -0.69 and -0.5618 values between -0.56 and -0.4362 values between -0.43 and -0.31254 values between -0.31 and -0.18873 values between -0.18 and -0.051793 values between -0.05 and 0.08696 values between 0.08 and 0.20159 values between 0.20 and 0.3325 values between 0.33 and 0.464 values between 0.46 and 0.593 values between 0.59 and 0.711 values between 0.71 and 0.842 values between 0.84 and 0.971 values between 1.48 and 1.601 values between 1.86 and 1.991 values between 2.11 and 2.241 values between 2.37 and 2.501 values between 3.13 and 3.262 values between 3.51 and 3.64offnegativepositiveoften

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. մ
  2. み
  3. veys
  4. 什么呢
  5. どう
  6. änn
  7. icators
  8. experience
  9. �
  10. 你看

Suppresses

  1. enville
  2. ственно
  3. 后果
  4. disadv
  5. idar
  6. 独角
  7. dra
  8. =&
  9. .rawValue
  10. setResult