Neuron Explorer

Qwen3-0.6B

unclear: mixed mid-sentence words

A neuron in layer 18 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

The strongest examples are scattered across many unrelated words, such as 'sign', 'all', 'frequently' and 'things', with no obvious shared meaning. Turning it up made the model's text wander and repeat, so this neuron is hard to pin down.

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

    essential vitamins and nutrients you need to lead an active, healthy lifestyle. While seniors may need to eat less than younger individuals, losing your appetite altogether could be a sign of a more serious medical condition.⏎What Causes Loss of Appetite in Seniors

  2. #2 peak

    After May 1, 2017, this browser will no longer be supported and you may experience issues or not be able to access the site at all. We highly recommend that you upgrade to a current version of

  3. #3 peak

    begin.⏎GCF and LCM by Prime Factorization⏎A straight forward way to find the greatest common factor (GCF) and least common multiple (LCM) using prime factorization. Article includes examples

  4. #4 peak

    may be an indication of a more serious problem.⏎One of the most common causes of frequent nose bleeds is dry air. Many people get nose bleeds more frequently in the winter, when the heat is on and the air is dryer. The

  5. #5 peak

    a future of happiness and success. However, many parents believe their child’s education begins on the first day of kindergarten. That is not true! Children learn new things every day, starting on the day they are born. In fact

  6. #6 peak

    converts rotary motion into linear motion.⏎The ‘lead’ of a Lead Screw is the distance which a Lead Screw Nut will move during one 360° rotation of the screw.

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 to do the same as the same as I have to do the same as what I can do... what does that mean, as a person to the same as what what the same as what the same as what the same... 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

    high-end models are capable of 1000, 1600 or even 2000 DPI tracking speeds.⏎The Personal System/2 or PS/2 was the designation for IBM's second generation of personal computers.

  2. #2 peak

    course of a network engineers career. A port number is the logical address of each application or process that uses a network or the internet to communicate.⏎User datagram protocol udp is a protocol that sends a datagram in its entirety from the host to

  3. #3 peak

    riculums, laboratories and teacher training. President Clinton's education legislation passed earlier this year by Congress, the Goals 2000: Educate America Act, sets a goal of making public school students No.

  4. #4 peak

    rot Loriculus flosculus and Yellow-crested Cockatoo Cacatua sulphurea.⏎Kenya’s Kikuyu Escarpment Forest IBA is home to a rich avifauna, characteristic of the country

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. L14:917 suppresses
  2. L12:336 excites
  3. L15:675 suppresses
  4. L14:169 excites
  5. L11:18 suppresses
  6. L14:348 excites
  7. L16:314 suppresses
  8. L16:928 excites
  9. L14:251 excites
  10. L13:132 suppresses "of" after a noun
  11. L15:286 suppresses
  12. L11:31 excites
  13. L16:2227 excites
  14. L17:694 excites
  15. L15:316 excites
  16. L9:7 suppresses
  17. L14:1091 suppresses
  18. L9:825 suppresses
  19. L10:1609 excites
  20. L9:584 excites
  21. L15:487 excites
  22. L14:126 excites
  23. L6:388 suppresses
  24. L5:1053 excites

Feeds later layers

  1. L19:148 excites
  2. L19:185 suppresses
  3. L20:12 suppresses
  4. L19:828 excites
  5. L25:102 excites
  6. L20:2366 excites
  7. L21:226 excites
  8. L19:1122 excites
  9. L19:1159 excites
  10. L20:1050 weakly excites
  11. L20:146 weakly excites
  12. L25:1310 weakly excites
  13. L19:479 weakly excites
  14. L19:2587 weakly excites
  15. L20:1904 weakly excites
  16. L21:81 weakly excites
  17. L20:245 weakly suppresses
  18. L19:1794 weakly excites
  19. L20:2873 weakly excites
  20. L25:120 weakly suppresses
  21. L20:71 weakly suppresses
  22. L19:2307 weakly excites
  23. L19:499 weakly excites
  24. L19:2988 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.

2 values between -8.53 and -8.131 values between -7.73 and -7.332 values between -7.33 and -6.935 values between -6.93 and -6.533 values between -6.53 and -6.136 values between -6.13 and -5.737 values between -5.73 and -5.337 values between -5.33 and -4.9318 values between -4.93 and -4.5317 values between -4.53 and -4.1336 values between -4.13 and -3.7344 values between -3.73 and -3.3349 values between -3.33 and -2.9348 values between -2.93 and -2.5360 values between -2.53 and -2.1377 values between -2.13 and -1.73170 values between -1.73 and -1.33265 values between -1.33 and -0.93414 values between -0.93 and -0.53454 values between -0.53 and -0.13481 values between -0.13 and 0.27419 values between 0.27 and 0.67283 values between 0.67 and 1.07249 values between 1.07 and 1.47212 values between 1.47 and 1.87163 values between 1.87 and 2.27129 values between 2.27 and 2.6799 values between 2.67 and 3.0787 values between 3.07 and 3.4742 values between 3.47 and 3.8724 values between 3.87 and 4.2712 values between 4.27 and 4.679 values between 4.67 and 5.074 values between 5.07 and 5.472 values between 5.47 and 5.873 values between 5.87 and 6.273 values between 6.27 and 6.671 values between 6.67 and 7.071 values between 7.07 and 7.47offnegativepositiveoften

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. 却又
  2. 及其
  3. 乃至
  4. elseif
  5. 及相关
  6. antor
  7. than
  8. 不然
  9. 的衣服
  10. 不已

Suppresses

  1. agen
  2. arians
  3. imators
  4. что
  5. oproject
  6. oom
  7. chợ
  8. lạc
  9. aland
  10. >O