Neuron Explorer

Qwen3-0.6B

middle digit of years in the 2000s and 2010s

A neuron in layer 18 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 responds to the number 2 in the middle of a year like 2020, 2021 or 2010, and it is most active there. It is less active, and sometimes negative, on the 1 in years like 2012, 2016 or 2019. 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: high. 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

    weather conditions and market factors.⏎The most recent report demonstrates cotton’s significant improvement in three of the five KPIs from 2010–2020 compared to the previous decade.⏎Irrigation water

  2. #2 peak

    still grew by an average of 200 people per day for the last decade, despite breaking its own heat-related records in repeated succession in 2020, including number of days over 110 degrees. To save lives

  3. #3 peak

    identified that the impact of religious education on Taiwanese students' civic values was low, with the main focus being personal and spiritual life development. Copyright © 2021 selection and editorial matter, Kerry J. Kennedy and John Chi-Kin Lee

  4. #4 peak

    Education, May 2019<|endoftext|>Over 2,000 Nigerian Farmers Reaping the Benefits of BiotechnologyJune 2, 2021⏎In 2019, six African countries have adopted biotech

  5. #5 peak

    mitigation strategies are tracked using the same technology, with 87 percent of the population currently residing in areas with approved mitigation plans (FEMA, 2020).⏎Emergency preparedness is the actions taken in advance that develop operational capabilities,

  6. #6 peak

    English as a Second Language examination and their teachers. All of the assessments and their instructions are based on the exemplar Core IGCSE paper from 2020 onwards and reflect recent changes to the structure and content of the paper. The

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 do not have to make the other day the only day I don't do something different that I do not do in the other day I do not do the other day I don't do the other day I do not do the other day

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

    more about tech gadgets that will protect your home against air pollutants<|endoftext|>Active and Passive Voice – Tips for Content Writing⏎February 22nd, 2012 by Donna Mathews⏎The use of active voice is important while writing online

  2. #2 peak

    HSC Rajasthan Board exam here: HSC Study Tips to Crack HSC Exams.⏎We had also shared Important Questions Bank for HSC Examination 2016 and students have really appreciated it and showered us with love last year.⏎

  3. #3 peak

    by the International Energy Agency (IEA) assesses a full range of energy technologies and sectors to provide a definitive snapshot of clean energy progress in 2019. As per the report, only 6 out of 46 technologies

  4. #4 peak

    oli are tiny air sacs within the lungs where the exchange of oxygen and carbon dioxide takes place.⏎Last Editorial Review: 6/14/2012⏎Back to MedTerms online medical dictionary A-Z List⏎Need help identifying

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. L8:457 excites year digits in copyright and dates
  2. L14:1700 excites
  3. L8:1185 suppresses
  4. L9:540 suppresses
  5. L10:1980 weakly excites
  6. L17:1476 weakly excites
  7. L8:3070 weakly excites
  8. L8:198 weakly suppresses
  9. L17:1964 weakly excites
  10. L10:2738 weakly excites
  11. L17:1255 weakly suppresses
  12. L3:438 weakly suppresses
  13. L2:1465 weakly excites
  14. L10:1142 weakly suppresses
  15. L5:100 weakly excites
  16. L17:1264 weakly excites
  17. L8:2473 weakly suppresses
  18. L16:626 weakly suppresses
  19. L0:951 weakly excites
  20. L8:23 weakly excites
  21. L8:224 weakly suppresses
  22. L4:2298 weakly suppresses
  23. L9:1791 weakly suppresses
  24. L10:2644 weakly suppresses

Feeds later layers

  1. L26:1875 weakly excites
  2. L26:1461 weakly excites
  3. L20:1124 weakly suppresses
  4. L20:1774 weakly excites
  5. L25:1651 weakly excites
  6. L25:1474 weakly suppresses
  7. L23:2262 weakly excites
  8. L21:1305 weakly excites
  9. L24:724 weakly excites
  10. L24:1575 weakly excites
  11. L23:2261 weakly excites
  12. L19:2182 weakly suppresses
  13. L21:2748 weakly excites
  14. L22:1180 weakly excites
  15. L27:697 weakly suppresses
  16. L20:2138 weakly suppresses
  17. L24:3 weakly suppresses
  18. L26:910 weakly excites
  19. L23:1882 weakly excites
  20. L27:1087 weakly suppresses
  21. L27:516 weakly suppresses
  22. L24:1153 weakly excites
  23. L19:2514 weakly suppresses
  24. L27:1596 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.

3 values between -8.10 and -7.281 values between -7.28 and -6.452 values between -5.62 and -4.801 values between -4.80 and -3.972 values between -3.14 and -2.326 values between -2.32 and -1.4937 values between -1.49 and -0.663337 values between -0.66 and 0.16491 values between 0.16 and 0.997 values between 0.99 and 1.827 values between 1.82 and 2.644 values between 2.64 and 3.473 values between 3.47 and 4.291 values between 4.29 and 5.124 values between 5.12 and 5.951 values between 6.77 and 7.601 values between 24.13 and 24.95offnegativepositiveoften

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. ury
  2. ybrid
  3. 垡
  4. nadzieję
  5. hma
  6. [r
  7. stole
  8. 阂
  9. eyed
  10. odash

Suppresses

  1. unta
  2. 十七条
  3. 득
  4. tuyến
  5. plet
  6. ограм
  7. billion
  8. >tag
  9. 十五条
  10. ulle