Neuron Explorer

Qwen3-0.6B

'with' after a comma in news clauses

A neuron in layer 8 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 fires on the word "with" when it comes after a comma and adds a follow-on description, as in "the debate has grown, with some people arguing..." Its negative side fires on words like "Next" and "Continue" in step-by-step instructions. Turning it up gave outputs with no obvious link to this pattern.

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

    Very few examples of stone lids have been recorded, suggesting that the jars were most likely capped with perishable material.⏎The function of the jars is still debated, with some archaeologists suggesting that they were prehistoric mortuary vessels, evident by the

  2. #2 peak

    doctors to provide the best possible care to our patients.”⏎Despite the expansion of telemedicine, studies of its effect on the quality of medical care remain scarce, with publications mostly limited to anecdotal reports or issues of technological feasibility and its potential to

  3. #3 peak

    surged dramatically—thanks largely to technological advances such as high-volume hydraulic fracturing, more commonly known as “fracking.” This rapid increase has generated widespread debate, with proponents touting economic and energy-security benefits and opponents highlighting the environmental and social risks

  4. #4 peak

    over the Pacific from an observation plane.⏎The European Space Agency's first cargo mission to the international space station ended in a spectacular fireworks show today, with the fiery re-entry of the unmanned Jules Verne ATV spaceship over the South

  5. #5 peak

    ak on Facebook|<|endoftext|>What State Is It Legal to Be Married to More than One Person⏎The legal status of polygamy varies from country to country, with each country prohibiting, accepting or promoting polygamy. In countries that accept or

  6. #6 peak

    urges Muslims to do, but one of the most controversial is the capture of non-Muslim women. This has been a source of debate and controversy for many years, with some people arguing that it is a barbaric practice that should be stopped, and

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 name of a property. The first time, I was in a room with you. You can do something. Now, let's say the first version. Now, let's say the first version. It is clear from the given data

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

    ) and type wireless. Click on Manage wireless networks.⏎- Click on Add to add a network.⏎- Click on Create an ad hoc network.⏎- Click on Next.⏎- Enter a name for your network and configure the security options. Click on

  2. #2 peak

    , there may be many other reasons of numeral conversion. This converter is able to easily convert any Arabic to Roman numeral for you.⏎It’s easy to use a converter for all types of users. You can even gift it to your kids so that

  3. #3 peak

    Browser and type “”http://Server-IP/drupal”” which will open below screen to configure Drupal. Select ‘Standard‘, click on Save and Continue.⏎Choose language, click on Save and Continue⏎Fill up the details like database

  4. #4 peak

    and click on Save and Continue.⏎Drupal installation has been completed.⏎Login in Drupal Site administration or Dashboard<|endoftext|>-

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. L7:1450 suppresses
  2. L4:1424 weakly excites
  3. L2:2899 weakly excites
  4. L6:725 weakly suppresses
  5. L7:1566 weakly suppresses
  6. L1:1246 weakly suppresses "word" after a language origin phrase
  7. L6:665 weakly excites
  8. L2:1343 weakly suppresses
  9. L2:2390 weakly excites
  10. L3:2967 weakly suppresses
  11. L5:2622 weakly suppresses
  12. L5:2979 weakly excites
  13. L1:1861 weakly suppresses
  14. L2:1201 weakly excites
  15. L1:2035 weakly suppresses
  16. L0:1591 weakly suppresses
  17. L6:2280 weakly excites
  18. L3:517 weakly suppresses
  19. L4:1889 weakly excites
  20. L5:117 weakly excites
  21. L4:589 weakly suppresses
  22. L1:2561 weakly excites
  23. L7:2691 weakly excites
  24. L5:284 weakly excites

Feeds later layers

  1. L9:2469 suppresses
  2. L12:1807 excites
  3. L13:1397 weakly excites
  4. L10:2394 weakly suppresses
  5. L10:1499 weakly suppresses
  6. L11:1075 weakly excites
  7. L12:1422 weakly suppresses
  8. L17:2465 weakly suppresses
  9. L9:1816 weakly suppresses
  10. L13:1674 weakly suppresses
  11. L11:2050 weakly suppresses
  12. L16:2845 weakly suppresses
  13. L13:1349 weakly suppresses
  14. L10:1258 weakly suppresses
  15. L11:2612 weakly excites
  16. L22:2841 weakly suppresses
  17. L10:1694 weakly suppresses
  18. L23:856 weakly suppresses
  19. L15:2960 weakly suppresses
  20. L9:3062 weakly excites
  21. L25:724 weakly suppresses
  22. L12:2287 weakly excites
  23. L10:1942 weakly suppresses
  24. L17:2657 weakly 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.

4 values between -0.62 and -0.5312 values between -0.53 and -0.4444 values between -0.44 and -0.34113 values between -0.34 and -0.25355 values between -0.25 and -0.16791 values between -0.16 and -0.071178 values between -0.07 and 0.03785 values between 0.03 and 0.12408 values between 0.12 and 0.21145 values between 0.21 and 0.3145 values between 0.31 and 0.4012 values between 0.40 and 0.496 values between 0.49 and 0.595 values between 0.59 and 0.681 values between 0.86 and 0.961 values between 1.79 and 1.891 values between 1.98 and 2.071 values between 2.07 and 2.171 values between 3.00 and 3.09offnegativepositiveoften

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. offsetof
  2. 礌
  3. оригинал
  4. ALTH
  5. 陣
  6. Providence
  7. fascinated
  8. papers
  9. flare
  10. 몰

Suppresses

  1. 柴
  2. unchecked
  3. ystate
  4. ivate
  5. ulations
  6. <Input
  7. utto
  8. Add
  9. ursions
  10. ypad