Neuron Explorer

Qwen3-0.6B

'to' or 'for' opening a purpose phrase

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 responds to small connecting words like 'to' or 'for' that introduce why something is done, such as 'enough light to get the shot'. It is strongly switched off on similar 'to' and 'so' phrases, so the neuron seems to care about the exact kind of purpose phrase, not just the word.

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

    a foot cushion or ideal camera rest when doing wildlife photography from your vehicle. (Really!)⏎Keep the HOTBOX away from open flame or fire. The heat required to cook the food is generated conventionally with electricity, gas, fire or paraff

  2. #2 peak

    tle Park stands today, to protect settlers from Native American attacks.⏎The steamboat era launched Cincinnati as a major

  3. #3 peak

    of sulfur in it is called “sour,” as opposed to “sweet.”) If enough of these contaminants collect in a steel pipe, they’ll work together to eat

  4. #4 peak

    .⏎Constant Vigilance for cancer recurrence has its own downsides.⏎After successful cancer treatment, being on look out for symptoms constantly wrecks your quality of life for minimal benefit in most cases.⏎Many trials of intensive screening shown no benefit for early

  5. #5 peak

    ’s Marine Protected Area Network can be found here.<|endoftext|>Origin and Etymology of kayak⏎First Known Use: 1757⏎KAYAK Defined for English Language Learners⏎Definition of kayak for English Language Learners⏎: a

  6. #6 peak

    afternoon. The sun was setting and there was just enough light to get the shot done. DeMille looked over the panorama, saw that all

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 can't be sure that I can be sure that I can be sure that I can be sure that I can be sure... the same as I can be sure that I can be sure that I can be sure... the same as I

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

    at war – the process of capturing the beasts is recorded in several Roman mosaics.⏎The cookie settings on this website are set to "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing

  2. #2 peak

    should be aware of the risk of herpes in metzitzah b’peh, and should inquire in advance whether direct orogenital suction will be performed so the practice can be avoided, the CDC researchers said.⏎"Oral contact with

  3. #3 peak

    enough smoke detectors in your house. Position them inside or just outside bedrooms, in the basement and garage, and at least 10 feet from your kitchen (to avoid false alarms).⏎- If you hear the detector go off and there’s no

  4. #4 peak

    organ of the plant which is designed for photosynthesis. Leaves collect carbon dioxide from the air and energy from the sun. They tend to be flat and thin to ensure

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. L9:1890 strongly excites
  2. L9:583 suppresses
  3. L6:2000 suppresses
  4. L6:1395 excites
  5. L7:1584 suppresses
  6. L11:896 excites comma closing an introductory phrase
  7. L8:796 excites
  8. L8:2824 suppresses
  9. L6:2880 excites
  10. L7:1828 excites
  11. L4:1035 suppresses 'for' after verbs like 'fall' or 'click'
  12. L10:1908 weakly suppresses 'for' after study resource nouns
  13. L8:2118 weakly excites
  14. L7:1673 weakly suppresses
  15. L5:1890 weakly excites
  16. L4:328 weakly suppresses 'to' after 'so as' and infinitive cues
  17. L10:2003 weakly excites
  18. L6:2897 weakly suppresses
  19. L1:496 weakly suppresses
  20. L11:125 weakly excites
  21. L0:1176 weakly excites
  22. L0:1006 weakly suppresses
  23. L1:1341 weakly suppresses
  24. L7:1234 weakly suppresses

Feeds later layers

  1. L17:2008 suppresses
  2. L15:1543 suppresses
  3. L20:922 suppresses
  4. L15:1312 suppresses
  5. L18:290 excites
  6. L14:869 suppresses
  7. L16:1477 suppresses
  8. L22:1806 suppresses "to" introducing a purpose clause
  9. L17:2856 suppresses
  10. L14:1308 suppresses
  11. L19:2738 suppresses
  12. L16:2685 suppresses
  13. L15:2536 suppresses
  14. L16:2933 suppresses
  15. L14:1158 suppresses 'to' starting a purpose clause
  16. L13:678 suppresses
  17. L14:2149 suppresses
  18. L17:2939 suppresses
  19. L16:2961 suppresses
  20. L24:715 suppresses
  21. L15:1943 excites
  22. L18:2900 suppresses
  23. L17:1347 suppresses
  24. L14:628 suppresses end of document boundary token

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 -6.36 and -6.181 values between -4.76 and -4.582 values between -4.58 and -4.401 values between -4.22 and -4.052 values between -3.87 and -3.691 values between -3.33 and -3.161 values between -3.16 and -2.981 values between -2.98 and -2.803 values between -2.62 and -2.452 values between -2.45 and -2.274 values between -2.27 and -2.092 values between -2.09 and -1.911 values between -1.91 and -1.743 values between -1.56 and -1.382 values between -1.38 and -1.204 values between -1.20 and -1.0210 values between -1.02 and -0.8518 values between -0.85 and -0.6723 values between -0.67 and -0.4952 values between -0.49 and -0.31296 values between -0.31 and -0.141584 values between -0.14 and 0.041356 values between 0.04 and 0.22426 values between 0.22 and 0.4098 values between 0.40 and 0.5814 values between 0.58 and 0.75offnegativepositiveoften

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. пуст
  2. WithValue
  3. offered
  4. 作出了
  5. OPSIS
  6. 怎么做
  7. (Void
  8. ivalent
  9. 拿出来
  10. 出现在

Suppresses

  1. ensures
  2. prevent
  3. readability
  4. 防止
  5. imm
  6. 确保
  7. 以便
  8. blade
  9. catch
  10. 保证