Neuron Explorer

Qwen3-0.6B

'over' meaning time or duration

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 lights up on the word "over" when it means a span of time or a finished process, as in "over 700 years ago", "over the years" or "over and over". It is quieter on the opposite side, where it switches off for sentences that begin with "So". Turning it up did not produce a clear theme in the written continuations.

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

    of Giza have been the objects of many conspiracies over the years. Many people believe that the Pyramids are a link to an ancient

  2. #2 peak

    wondering? Well, think about analog watches. They keep time thanks to a mechanical architecture that siphons off kinetic energy from you moving about, and stores it over time. The team wondered if they could borrow principles like this in their own marvelous

  3. #3 peak

    Detoxification – Part 1⏎Updated: Sep 24, 2018⏎Detoxification – you’ve heard this word many times over the last few years, but unless you have actually participated in a detoxification program

  4. #4 peak

    with print. Get a newspaper delivered. Get her a library card and make the library a regular stop, like the grocery store and the dry cleaner. And get over your view of what "proper" book reading looks like — fiction, nonfiction

  5. #5 peak

    /labs2.htm)View an example of a Dataforth test report that ships with each module.Remember, our Application Engineers can assist you with signal conditioner selection over the phone or via fax and email. Call us at our

  6. #6 peak

    resolved, if we just try hard enough not to think about them, as if it were a choice to be impacted by our worst experiences, to relive them over and over again, and to react as if we were under threat.⏎Let’s

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: The way in the box, the way in the light, the way in the theme, and the like, that has been for a family, and for a person, and on the like, with the family, with the family, in

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

    Use. The complexity and uncertainty of a Fair Use analysis make it both risky and cumbersome to apply to small projects involving borrowing Internet materials<|endoftext|>3D Television Technology⏎So, you’ve seen it on the news, read about it in the papers and

  2. #2 peak

    But what exactly is Diabetes? There are a lot of myths and misunderstandings surrounding the disease, particularly when it comes to type 1 versus type 2.⏎So let’s start with

  3. #3 peak

    The Sound of Silence” to “Bridge Over Troubled Water”, their songs act<|endoftext|>Read Music Notes⏎How to Read All the Music Notes -- With Ease⏎So

  4. #4 peak

    ?⏎As one of the lesser known grains out there, buckwheat (and what it is good for) can often be a point of confusion for many. So, before we get any further – we first think that it would be a good

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. L17:213 strongly suppresses
  2. L16:177 strongly excites zero digit inside dates and codes
  3. L2:1795 strongly suppresses
  4. L5:1546 strongly suppresses
  5. L0:44 strongly suppresses
  6. L10:1052 strongly suppresses
  7. L1:15 strongly suppresses
  8. L4:498 strongly suppresses
  9. L1:178 suppresses
  10. L0:1241 excites
  11. L0:1906 suppresses
  12. L7:1337 excites 'of' inside organisation and product names
  13. L2:1463 excites
  14. L16:164 excites
  15. L17:1046 suppresses
  16. L16:827 suppresses
  17. L0:2302 excites
  18. L15:381 excites
  19. L17:337 suppresses
  20. L7:342 excites
  21. L15:1616 weakly excites word fragments mid-word
  22. L16:1066 weakly suppresses
  23. L15:1404 weakly suppresses
  24. L15:1587 weakly excites

Feeds later layers

  1. L20:392 strongly suppresses
  2. L21:381 strongly suppresses zeros inside long number codes
  3. L20:2206 strongly suppresses
  4. L19:1370 strongly suppresses
  5. L20:3008 strongly suppresses
  6. L20:2261 strongly suppresses
  7. L21:2826 strongly suppresses
  8. L20:2363 suppresses
  9. L23:2981 suppresses
  10. L23:1324 suppresses
  11. L19:843 suppresses
  12. L19:1696 suppresses
  13. L19:906 suppresses
  14. L20:1374 suppresses
  15. L21:1052 suppresses mid-word 'electro' and 'ro' fragments
  16. L19:798 excites
  17. L27:1585 suppresses
  18. L21:2302 suppresses
  19. L22:2365 suppresses
  20. L26:1290 excites
  21. L27:2149 suppresses
  22. L19:390 excites
  23. L26:2846 suppresses
  24. L19:316 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 -15.20 and -14.472 values between -14.47 and -13.741 values between -13.00 and -12.273 values between -12.27 and -11.541 values between -11.54 and -10.804 values between -10.80 and -10.074 values between -10.07 and -9.342 values between -9.34 and -8.6014 values between -8.60 and -7.8711 values between -7.87 and -7.1420 values between -7.14 and -6.4035 values between -6.40 and -5.6744 values between -5.67 and -4.9466 values between -4.94 and -4.20112 values between -4.20 and -3.47136 values between -3.47 and -2.74155 values between -2.74 and -2.00175 values between -2.00 and -1.27239 values between -1.27 and -0.54365 values between -0.54 and 0.201437 values between 0.20 and 0.931072 values between 0.93 and 1.675 values between 1.67 and 2.401 values between 2.40 and 3.131 values between 3.13 and 3.871 values between 9.00 and 9.731 values between 13.40 and 14.13offnegativepositiveoften

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. faculties
  2. Hundreds
  3. insula
  4. alore
  5. pylint
  6. uent
  7. ▇
  8. /xhtml
  9. 昂
  10. ⏎ ⏎

Suppresses

  1. ongan
  2. iginal
  3. entimes
  4. ставить
  5. ductive
  6. Letter
  7. WithTag
  8. aurus
  9. sid
  10. petition