Neuron Explorer

Qwen3-0.6B

function words before an expected phrase

A neuron in layer 16 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

This neuron fires on small connecting words that set up a phrase still to come, like 'of' in 'think of', 'on' in 'based on', or 'the' after 'took'. Turning it up made the model's text wander, with no clear theme.

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

    antages and disadvantages of using Sunpower solar panels⏎The use of solar energy is becoming increasingly popular across the globe as the technology continues to develop and prices drop. One of the most popular types of solar energy is Sunpower solar panels. Sunpower

  2. #2 peak

    separation anxiety. When left alone for prolonged periods of time, dogs will begin to feel lonely and frightened. This can cause them to act out in many ways, including

  3. #3 peak

    in the garment industry has transformed cultural expectations of women and girls.<|endoftext|>Microcontroller vs FPGA: Understanding the Key Differences⏎When it comes to embedded systems design, two technologies that often come up are microcontrollers and field-programmable gate arrays (

  4. #4 peak

    24 hours.⏎3D printed valves⏎Once the Brescia hospital realised its regular valve supplier could not produce the valves on time, it was forced to think of alternative solutions as it had run out of breathing tubes for the intensive care machine.

  5. #5 peak

    that supposedly must have existed. Evolutionists needed evidence to back up the supposed transition of vertebrates from the sea to dry land. For that reason, they took the fossil coelacanth, whose anatomy they believed was ideally suited to this scenario

  6. #6 peak

    worldwide,” Waring says. “To combat global warming, we’re understandably going to make our buildings more sustainable. But in that process, we need to think of sustainability as both reducing energy demand and protecting the health of the building occupants.”⏎W

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 reason for the difference between the current time, the current time, and the difference between the current time. This is what you are saying, but I think the problem with this, the difference between the current time and the current time,

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

    by By Dr. Carolyn K. Stratz⏎Q: My kids get their vaccinations - what about me?⏎A: Back-to-school time doesn't just mean immunizations for

  2. #2 peak

    was found in River Rio Cristalino in the southern Amazonas. It's nearest relative in the aquaristic is Hygrophila but Staurogyne is different with its marked compact, low and bushy stature (5-1

  3. #3 peak

    to come,” but “this is not a disease that the general public should be worried about. It’s not Covid or other fast-spreading diseases.” Monkeypox belongs to the same family as smallpox, which killed millions of people

  4. #4 peak

    . Other tools on the Voyant site are Word links (creates a mind map of words) and a scatter plot (for word frequencies). http://voyant-tools.org/⏎Wordle: Wordle is a free web<|endoftext|>His wife once threatened

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. L15:245 suppresses
  2. L13:2538 excites
  3. L15:2963 excites
  4. L15:327 excites
  5. L14:2957 suppresses
  6. L13:801 suppresses
  7. L13:298 suppresses
  8. L13:1018 suppresses
  9. L13:1611 suppresses
  10. L14:219 excites
  11. L15:2476 suppresses
  12. L13:1347 excites
  13. L15:707 suppresses
  14. L15:1734 excites
  15. L15:2577 suppresses
  16. L14:2069 suppresses
  17. L13:2312 excites
  18. L13:1910 excites
  19. L13:1407 suppresses
  20. L15:159 suppresses
  21. L15:2641 excites
  22. L15:268 suppresses
  23. L7:2470 suppresses
  24. L14:891 excites

Feeds later layers

  1. L17:2175 excites
  2. L17:1687 excites
  3. L17:2094 suppresses
  4. L17:1151 excites
  5. L17:1624 excites
  6. L17:2851 excites
  7. L17:1098 suppresses
  8. L17:1329 excites
  9. L17:655 weakly excites
  10. L19:431 weakly excites
  11. L18:2291 weakly suppresses
  12. L17:2173 weakly suppresses
  13. L23:1339 weakly excites
  14. L19:1916 weakly excites
  15. L17:1915 weakly excites
  16. L17:1975 weakly excites
  17. L18:1886 weakly suppresses
  18. L19:823 weakly excites
  19. L17:1124 weakly suppresses
  20. L17:1240 weakly excites
  21. L17:1510 weakly suppresses
  22. L18:941 weakly suppresses
  23. L24:470 weakly suppresses
  24. L18:2275 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 -2.03 and -1.922 values between -1.60 and -1.491 values between -1.49 and -1.381 values between -1.38 and -1.276 values between -1.27 and -1.164 values between -1.16 and -1.056 values between -1.05 and -0.944 values between -0.94 and -0.8319 values between -0.83 and -0.7220 values between -0.72 and -0.6135 values between -0.61 and -0.5058 values between -0.50 and -0.40101 values between -0.40 and -0.29196 values between -0.29 and -0.18536 values between -0.18 and -0.071792 values between -0.07 and 0.04703 values between 0.04 and 0.15206 values between 0.15 and 0.2685 values between 0.26 and 0.3745 values between 0.37 and 0.4830 values between 0.48 and 0.598 values between 0.59 and 0.7018 values between 0.70 and 0.814 values between 0.81 and 0.929 values between 0.92 and 1.037 values between 1.03 and 1.133 values between 1.13 and 1.242 values between 1.24 and 1.354 values between 1.46 and 1.571 values between 1.79 and 1.901 values between 2.23 and 2.34offnegativepositiveoften

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. rolled
  2. acular
  3. 山区
  4. 爱人
  5. acles
  6. 旅行
  7. avorite
  8. コンテンツ
  9. ipur
  10. @PostMapping

Suppresses

  1. kicking
  2. flu
  3. grub
  4. arm
  5. redirects
  6. obao
  7. retaining
  8. BLE
  9. epam
  10. grabbing