Neuron Explorer

Qwen3-0.6B

unclear: mid-phrase words before a noun

A neuron in layer 14 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 a word in the middle of a sentence, often the word just before a noun like 'time', 'song' or 'music'. The pattern is loose, and turning it up just made the model's text wander off topic.

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

    him/her and will learn to crawl, stand and eventually walk.⏎Why is tummy time important?⏎Infants spend a lot of time lying on their back, especially a few weeks after being born. If they’re

  2. #2 peak

    the South in the 19th century with the emergence of the Atlantic Slave trade. It was during this time that thousands of Africans were moving to America with music influences from their native

  3. #3 peak

    with your melody, but, in most cases, they won't match up perfectly. The melody should usually take precedence over the lyrics, as the tune of the song is more

  4. #4 peak

    choice continues until the acquisition of another domicile of choice.⏎1. Domicile of choice: acquire when abandon previous domicile, arrive in the new domicile (must have some physical presence), and intend to remain in new domicile for

  5. #5 peak

    which is done on windy days, is also a memorial of this celebration.⏎Iranologists have two doubts about the timing of this celebration. The first group believes that this celebration has been held in Badruz since January, and within a week of its

  6. #6 peak

    scientists at Warwick pushed the limits of imaging the olympicene with their microscopy equipment they contacted the IBM scientists in an attempt to see whether they could use their technique for even greater resolution. The result? A beautifully rendered image of a single molecule

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 a 3D structure, a 2D structure, and a 4D structure, with the following properties: - The total volume is 1,000,000,000,00

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

    matters.⏎If a limited partner assumes an active role in management, that partner may lose his limited liability protection and may be treated as a general partner. For instance, if a limited partner negotiates a contract with a third party on behalf of the

  2. #2 peak

    with a possibly faulty mass air flow sensor.⏎Sometimes these symptoms may be due to other reasons, but you should still have your mass air flow sensor inspected if you experience at least a few of these symptoms that are listed below.⏎1) Lean Id

  3. #3 peak

    a copyright-protected document to tell the story. Another example of "fair use" is when a university professor quotes several sentences from a copyright-protected book in a review<|endoftext|>Sixty-three men from the parish lost their lives in the war.

  4. #4 peak

    understanding can lead to relief from pain, stress and tension, increase energy, improve coordination and enhance performance skills.⏎Study with Molly in Kansas City or bring her to you……anywhere on this beautiful planet.⏎Alexander Technique Choreography Dance Movement Performance<|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. L12:2214 excites
  2. L13:922 suppresses
  3. L13:671 excites unclear: maintenance, safety and process nouns
  4. L10:2177 excites
  5. L10:2018 suppresses
  6. L13:138 excites
  7. L11:2039 excites
  8. L13:1298 suppresses
  9. L11:663 excites
  10. L12:1489 excites
  11. L12:1757 suppresses
  12. L12:2236 suppresses
  13. L13:1915 excites
  14. L11:679 suppresses
  15. L10:1851 excites
  16. L13:832 excites
  17. L12:1586 excites
  18. L13:1717 excites
  19. L12:2565 weakly suppresses
  20. L13:662 weakly excites
  21. L13:2031 weakly excites
  22. L11:1649 weakly excites
  23. L13:2508 weakly excites
  24. L12:1571 weakly excites

Feeds later layers

  1. L15:1807 excites
  2. L17:2220 excites
  3. L15:1076 excites
  4. L16:2084 excites
  5. L15:2241 excites
  6. L16:1207 excites
  7. L16:713 excites
  8. L18:2515 excites
  9. L15:2096 suppresses
  10. L15:2838 excites
  11. L16:403 weakly excites
  12. L16:2426 weakly excites
  13. L16:1995 weakly excites
  14. L21:1080 weakly suppresses
  15. L15:2588 weakly excites
  16. L16:2753 weakly excites
  17. L18:2903 weakly excites
  18. L15:2816 weakly suppresses
  19. L19:700 weakly excites
  20. L15:2420 weakly excites
  21. L18:2797 weakly excites
  22. L19:2430 weakly excites
  23. L16:1342 weakly excites
  24. L18:3008 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 -2.08 and -2.001 values between -1.92 and -1.841 values between -1.84 and -1.772 values between -1.61 and -1.531 values between -1.53 and -1.452 values between -1.45 and -1.373 values between -1.37 and -1.293 values between -1.29 and -1.221 values between -1.22 and -1.141 values between -1.14 and -1.062 values between -1.06 and -0.985 values between -0.98 and -0.905 values between -0.90 and -0.824 values between -0.82 and -0.7410 values between -0.74 and -0.6719 values between -0.67 and -0.5921 values between -0.59 and -0.5137 values between -0.51 and -0.4350 values between -0.43 and -0.3578 values between -0.35 and -0.27127 values between -0.27 and -0.20247 values between -0.20 and -0.12617 values between -0.12 and -0.041671 values between -0.04 and 0.04566 values between 0.04 and 0.12212 values between 0.12 and 0.2091 values between 0.20 and 0.2846 values between 0.28 and 0.3630 values between 0.36 and 0.4318 values between 0.43 and 0.5113 values between 0.51 and 0.595 values between 0.59 and 0.674 values between 0.67 and 0.752 values between 0.75 and 0.835 values between 0.83 and 0.914 values between 0.91 and 0.983 values between 0.98 and 1.06offnegativepositiveoften

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. атор
  2. ادية
  3. ricular
  4. ETY
  5. atory
  6. atoria
  7. руч
  8. ickness
  9. ordable
  10. EGIN

Suppresses

  1. 咫
  2. maxi
  3. utas
  4. nextState
  5. 笑脸
  6. quila
  7. mates
  8. major
  9. knowledge
  10. 饱