Neuron Explorer

Qwen3-0.6B

comma after a 'before...' clause

A neuron in layer 12 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 comma that closes an opening phrase starting with "Before" or "Prior to", such as "Before making your ornaments, discuss". It is looking for the point where the main sentence is about to start. Turning it up did not produce a clear, related change in the text.

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

    consider the biblical story of the coming of Jesus.⏎The Star of Bethlehem⏎Read Matthew 2:1-12 with your kids.⏎Before making your ornaments, discuss:⏎- Where was Jesus born? (Matthew 2:1)⏎-

  2. #2 peak

    2 hours for seven days. The drops must be warmed to body temperature by putting the bottle in your hand for a few seconds before use. Before administering the drops, washing the affected ear(s) and drying them with a clean towel is necessary.

  3. #3 peak

    Question: How many days should Passover be observed? Please include any references to your response that is indicated in the Torah.⏎Answer: Before I answer your question, as a word of background, you need to keep in mind that Judaism, ever

  4. #4 peak

    and tribal law; other times you may be researching one or the other. This guide covers researching both areas of law.<|endoftext|>Before we may design a crystal filter it is necessary

  5. #5 peak

    the young boys are led out of seclusion and back into the community. Before festivities can begin, the head teacher (kahyuudi or kayudi) commissions a carver (

  6. #6 peak

    affected ear looks up when applying the drops. To widen the ear canal:⏎- Gently pull the earlobe back and down.⏎- Before using the ear drops, give the bottle a good shake to mix the medication.⏎- Apply the recommended number

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 need to do something and I have to do something, and then I am trying to do something, I am trying to write a paper, and then, I have to write a paper, and then, in this case, I have

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 relative upward and downward mobility — i.e. movement up or down the socio-economic ladder. In other words, are Americans born poor becoming rich and vice-versa?⏎Interestingly enough, geographic mobility doesn’t appear have a large effect on economic mobility

  2. #2 peak

    from the Space Sciences Department at NASA’s Marshall Space Flight Center in Huntsville, Ala., will launch a weather balloon to 100,000 feet about 2 a.m. either Aug. 12 or 13

  3. #3 peak

    or architect decomposes the system requirements into functional requirements for both the software and the hardware subsystems. Traceability information lets the development team track where each requirement is addressed in the design.⏎Poor allocation decisions can result in the software being expected to perform

  4. #4 peak

    all-loving creator who meets every human need. In Psalms we read, “Give thanks to the Lord, for he is good; his love endures forever. Let the redeemed of the Lord tell their story. . . . Let them

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. L8:2281 suppresses
  2. L6:2322 excites
  3. L10:2412 excites
  4. L11:2037 suppresses
  5. L7:2050 suppresses
  6. L10:463 suppresses
  7. L6:2303 suppresses
  8. L6:230 suppresses
  9. L5:2485 suppresses
  10. L7:962 excites
  11. L9:2052 excites 'because' and 'that' after a reason
  12. L4:1839 weakly suppresses
  13. L11:896 weakly suppresses comma closing an introductory phrase
  14. L10:1518 weakly excites
  15. L10:322 weakly excites
  16. L2:2160 weakly suppresses
  17. L6:1822 weakly suppresses
  18. L6:3024 weakly suppresses
  19. L8:170 weakly suppresses
  20. L8:447 weakly suppresses
  21. L10:3037 weakly suppresses
  22. L6:2754 weakly excites
  23. L5:2669 weakly excites
  24. L9:1334 weakly excites

Feeds later layers

  1. L14:957 excites
  2. L14:709 excites
  3. L15:2517 excites
  4. L15:2257 excites unclear: word stems in technical text
  5. L14:2512 suppresses
  6. L15:2338 suppresses
  7. L14:620 excites
  8. L14:576 excites
  9. L17:1225 suppresses
  10. L14:2392 suppresses
  11. L17:1156 excites
  12. L16:1984 weakly excites
  13. L19:544 weakly excites
  14. L16:1995 weakly excites
  15. L13:887 weakly suppresses
  16. L16:501 weakly suppresses
  17. L15:1369 weakly excites
  18. L25:3019 weakly excites
  19. L16:488 weakly excites
  20. L20:864 weakly excites
  21. L18:2296 weakly suppresses
  22. L20:796 weakly suppresses
  23. L15:456 weakly suppresses 'and' between two verbs
  24. L17:1076 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.

2 values between -1.62 and -1.501 values between -1.39 and -1.272 values between -1.16 and -1.042 values between -1.04 and -0.923 values between -0.92 and -0.817 values between -0.81 and -0.6914 values between -0.69 and -0.5821 values between -0.58 and -0.4641 values between -0.46 and -0.34116 values between -0.34 and -0.23339 values between -0.23 and -0.111587 values between -0.11 and 0.011238 values between 0.01 and 0.12304 values between 0.12 and 0.24109 values between 0.24 and 0.3544 values between 0.35 and 0.4719 values between 0.47 and 0.5912 values between 0.59 and 0.7014 values between 0.70 and 0.8212 values between 0.82 and 0.932 values between 0.93 and 1.055 values between 1.05 and 1.173 values between 1.17 and 1.282 values between 1.28 and 1.401 values between 1.40 and 1.511 values between 1.51 and 1.631 values between 1.75 and 1.862 values between 1.86 and 1.981 values between 2.21 and 2.331 values between 2.33 and 2.441 values between 2.56 and 2.671 values between 2.91 and 3.02offnegativepositiveoften

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. icense
  3. 的父亲
  4. tle
  5. prise
  6. (pol
  7. )prepare
  8. 却被
  9. /close
  10. 負

Suppresses

  1. ef
  2. otify
  3. evin
  4. uit
  5. ฉบ
  6. 八年
  7. ekt
  8. 做人
  9. ore
  10. ในการ