Neuron Explorer

Qwen3-0.6B

blank before a time or count

A neuron in layer 22 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 is active on the empty spot just before a number that tells how long something lasts, like 'up to [4] weeks', 'by [24] to 72 hours' or 'for [10] days'. It is most active when the model expects a quantity of days, weeks or months. Its negative side fires on the word 'to' or 'of' in ranges like 'one to two' or 'three out of four'. Turning it up produced numeric text such as clock times and measurements, so it seems to track counting.

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

    3 to 8 days, but small infants who have been infected may be contagious for up to 4 weeks.⏎In small infants RSV also causes mucus membrane damage and clear

  2. #2 peak

    you have certain other medical problems.⏎Alcohol withdrawal symptoms usually occur within 8 hours after the last drink, but can occur days later. Symptoms usually peak by 24 to 72 hours, but may go on for weeks.⏎Common

  3. #3 peak

    inal cavity.⏎Neurosurgical intervention may leave gas trapped within the skull, which again may expand at altitude. It is therefore advisable to avoid air travel for approximately 7 days following this type of procedure.⏎Ophthalmological procedures for retinal detachment

  4. #4 peak

    ⏎What to Do⏎- Setup materials as shown in the photo above. Note that the 4 styrofoam cups support the clear plastic dish and that 3 are upside-down

  5. #5 peak

    Chronic pelvic pain is medically defined as the persistent pain in the pelvic region, which may last for 3 months (non-cyclical pain) or for 6 months (cyclical pain). Four primary mechanisms lead to the generation of

  6. #6 peak

    in extreme circumstances perforation. Stretching intestinal or gastric mucosa may also result in haemorrhage.⏎To avoid such complications, travel should be avoided for 10 days following abdominal surgery. Following other procedures, such as colonoscopy

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: one of the key reasons that one can't just just have a general answer to the problem set in a form that can be easily to to get to set to one

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

    away about one-third of the lower root system to create compact roots.⏎- Place the bonsai in a new container.⏎- One year later, repeat steps one to three of this process.⏎Why Do I Need to Trim the Roots of My B

  2. #2 peak

    usually 10 to 14 days.⏎The adult female Anopheles mosquito can live up to a month, but typically does not live more than one to two weeks. The adult stage is when the female mosquito acts as malaria vector (

  3. #3 peak

    we were to do this Punnett square of this cross, we would end up with this. So basically, 75% of their children, three out of four, have at least one dominant allele, meaning that they're going to show

  4. #4 peak

    means that astronauts might potentially be able to harvest their own food in space—allowing them to travel to distant planets like Mars, which takes a grueling two and a half years to get to. While this announcement is exciting news, growing things

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. L21:175 strongly suppresses
  2. L16:372 excites
  3. L18:1006 suppresses
  4. L20:1998 suppresses
  5. L19:2929 suppresses
  6. L21:836 excites
  7. L12:752 excites
  8. L19:1921 excites
  9. L19:728 suppresses
  10. L19:1985 excites
  11. L20:2316 suppresses
  12. L19:298 excites
  13. L21:1284 excites
  14. L13:1083 excites
  15. L21:1647 excites
  16. L20:1011 suppresses
  17. L19:1260 excites
  18. L20:780 excites 'of' before a measured value
  19. L20:1001 suppresses
  20. L20:2136 excites
  21. L14:2494 suppresses
  22. L21:785 excites
  23. L17:2655 excites
  24. L20:875 excites

Feeds later layers

  1. L23:1264 strongly suppresses
  2. L25:572 strongly suppresses
  3. L24:2948 strongly suppresses
  4. L25:2915 strongly suppresses
  5. L26:1562 strongly suppresses
  6. L27:696 strongly suppresses
  7. L24:1406 suppresses
  8. L23:250 excites
  9. L23:198 excites
  10. L26:1083 suppresses
  11. L26:1617 suppresses
  12. L25:598 suppresses
  13. L26:1028 suppresses
  14. L27:238 excites
  15. L26:611 excites
  16. L26:667 excites
  17. L27:113 suppresses
  18. L27:48 suppresses
  19. L25:253 excites
  20. L25:921 suppresses
  21. L27:1422 suppresses
  22. L26:57 suppresses
  23. L24:197 excites
  24. L27:241 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 -19.63 and -18.181 values between -16.74 and -15.292 values between -15.29 and -13.851 values between -12.40 and -10.961 values between -10.96 and -9.514 values between -9.51 and -8.072 values between -8.07 and -6.623 values between -6.62 and -5.1811 values between -5.18 and -3.7434 values between -3.74 and -2.29120 values between -2.29 and -0.853345 values between -0.85 and 0.60373 values between 0.60 and 2.046 values between 2.04 and 3.492 values between 3.49 and 4.931 values between 4.93 and 6.381 values between 36.71 and 38.16offnegativepositiveoften

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. usra
  2. ello
  3. .nlm
  4. 第
  5. ainen
  6. nge
  7. ulators
  8. leigh
  9. かけて
  10. mund

Suppresses

  1. 四
  2. three
  3. 五
  4. four
  5. four
  6. five
  7. 七
  8. 六
  9. 八
  10. five