Neuron Explorer

Qwen3-0.6B

"is" after a sentence-opening "It"

A neuron in layer 19 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 "is" when it comes straight after "It" at the start of a sentence, as in "It is situated" or "It is estimated". It is turned off by "is" in other spots, such as after "question". Turning it up did not give a clear pattern in the writing.

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

    Shiva. Dhyansar Mahadev Temple is known for its vibrant festivals and celebrations. Mahashivratri is celebrated with great pomp here. It is celebrated with great enthusiasm by the local people. The local people believe that whatever is

  2. #2 peak

    contamination. These products are naturally greasy, making the lubrication unnecessary. Another advantage to choosing graphite for a bearing is that it is extremely temperature resistant. It is able to maintain its structure and shape even if exposed to temperatures as high as

  3. #3 peak

    jobs in government institutions like schools, colleges, courts etc.<|endoftext|>I hope you are all well and that you enjoyed some of your weekend in the sunshine. It is now time to spend a brand new week completing our learning jobs at home with our

  4. #4 peak

    is perfect and I am certain they will be excited to complete this! It is clever and helpful!”- Buyer⏎♥ “It was good to find something a

  5. #5 peak

    Mahadev Temple is located in Dhyansar village of Una district of Himachal Pradesh. It is situated at an altitude of about 1,500 meters above sea level

  6. #6 peak

    much shorter range than electromagnetic force and gravity force. Therefore, neutrinos are the most penetrating subatomic particles, capable of passing through Earth without any interaction. It is estimated neutrinos have interaction cross-sections about 20 orders of magnitude less

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: to get to a person to get a to what to what to what to what to what to what to what to what to what to what to what to what to what to what to what to what to what to what to what to what

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

    edges, but I guess there is some algorithm which is able to minimize the number of crossing edges. Can anyone help me?⏎Perhaps an "easier" question is: how can I determine whether two edges will be crossing? In graphs coordinates are

  2. #2 peak

    did not address is a comparison of human motion and more traditional machine motion. While humans are very flexible, they are often not very precise. A more interesting question is what is the optimal type of motion.⏎I'm with you Naperlou

  3. #3 peak

    , who spent years studying the coelacanth, described how many hid behind it as a lone piece of evidence:⏎One of the great problems of evolution has been to find anatomical links between the fishes and their land-invading descendants . .

  4. #4 peak

    munotherapy)⏎For people suffering severe allergic symptoms from allergic rhinitis, allergic asthma, conjunctivitis or stinging insect allergy, a long term solution is incredibly appealing. Allergy shots, also known as allergen immunotherapy, are

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. L16:466 strongly excites 'is' after a long subject phrase
  2. L17:227 suppresses
  3. L18:960 suppresses
  4. L14:827 suppresses open bracket after a value or citation
  5. L9:66 excites
  6. L4:20 suppresses
  7. L15:599 excites
  8. L4:24 excites
  9. L13:897 excites
  10. L8:137 suppresses
  11. L17:2353 excites
  12. L1:301 suppresses
  13. L0:207 excites
  14. L18:857 suppresses
  15. L13:652 suppresses
  16. L12:434 suppresses
  17. L6:858 excites
  18. L18:144 suppresses
  19. L14:149 weakly suppresses
  20. L1:1791 weakly suppresses
  21. L6:2237 weakly excites
  22. L18:93 weakly suppresses
  23. L17:882 weakly suppresses
  24. L3:76 weakly suppresses

Feeds later layers

  1. L21:2711 suppresses
  2. L20:2262 suppresses
  3. L20:2883 suppresses
  4. L21:81 excites
  5. L25:114 excites
  6. L20:2010 suppresses
  7. L20:1956 weakly suppresses
  8. L21:554 weakly excites
  9. L25:247 weakly suppresses
  10. L23:914 weakly suppresses
  11. L22:763 weakly excites
  12. L21:2277 weakly suppresses
  13. L21:918 weakly suppresses
  14. L20:2692 weakly suppresses
  15. L20:64 weakly excites
  16. L23:2361 weakly suppresses 'to' in 'to do with'
  17. L27:262 weakly suppresses
  18. L22:228 weakly excites
  19. L21:443 weakly excites
  20. L23:489 weakly excites
  21. L22:427 weakly excites
  22. L22:590 weakly suppresses
  23. L20:747 weakly excites
  24. L23:243 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.

1 values between -23.39 and -22.493 values between -16.21 and -15.311 values between -12.62 and -11.721 values between -10.83 and -9.931 values between -9.93 and -9.033 values between -9.03 and -8.134 values between -8.13 and -7.243 values between -7.24 and -6.346 values between -6.34 and -5.444 values between -5.44 and -4.5414 values between -4.54 and -3.6519 values between -3.65 and -2.7525 values between -2.75 and -1.8555 values between -1.85 and -0.951380 values between -0.95 and -0.062257 values between -0.06 and 0.8463 values between 0.84 and 1.7418 values between 1.74 and 2.6411 values between 2.64 and 3.5312 values between 3.53 and 4.4310 values between 4.43 and 5.335 values between 5.33 and 6.235 values between 6.23 and 7.121 values between 7.12 and 8.022 values between 8.02 and 8.923 values between 8.92 and 9.821 values between 11.61 and 12.51offnegativepositiveoften

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. -
  2. iable
  3. em
  4. 的一种
  5. 而
  6. ��
  7. 日报
  8. 直
  9. ES
  10. _

Suppresses

  1. lections
  2. annes
  3. imiento
  4. amientos
  5. aan
  6. aminer
  7. ания
  8. uctions
  9. vern
  10. anos