Neuron Explorer

Qwen3-0.6B

second half of multi-word names

A neuron in layer 2 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 the final word of a multi-word name or term, such as 'Organization' in Food and Agriculture Organization, 'Community' in European Coal and Steel Community, or 'Sea' in Coral Sea. It is strongly negative on words like 'attendance' and 'crash' in phrases like car crash. Turning it up did not give a clearly related change 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

    the phone at home and focus on your experience.⏎- Consider taking a walk in the woods. Researchers have learned that "forest bathing" can decrease stress hormones and increase proteins that fight cancer.⏎- Please see your

  2. #2 peak

    and fun games for children with autism into your visual supports too. To them they’re playing games, but you know that they’re getting some occupational therapy, speech activities and thinking skills thrown in. Games may also help children with autism engage more readily

  3. #3 peak

    to ratify. [iii] Successful European integration efforts of the time instead focused on economics. The EU’s forerunners, the European Coal and Steel Community, the European Atomic Energy Community, and the European Economic Community, were all established

  4. #4 peak

    came on 7 May 1942, when the US aircraft carriers USS Lexington and USS Yorktown faced three Japanese carriers in the Battle of the Coral Sea. During the battle, which lasted two days and was the first naval battle in

  5. #5 peak

    is a collection of ASL storytelling, poetry, works of art, and other creative works. It also consists of posts on literary aspects of ASL.⏎Speech language can convey sound effects in storytelling, whereas sign language can convey cinematic effects in storytelling

  6. #6 peak

    , focused on the empowerment of the rural and urban poor. Between 1976 and 2001, she worked with the Food and Agriculture Organization (FAO) of the United Nations to identify and support innovative NGO initiatives for

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 lot of time in the past, and I have been thinking about something that is a problem in my life." " I have been thinking about this for a long time, and I have been thinking about something that is a

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

    psychiatric – indicate that school attendance is not advisable. Additionally, homebound instruction may be used to supplement the classroom

  2. #2 peak

    an ordinary, unremarkable beginning to a life dependent on opioids. Someone might experience a routine surgery or have some kind of accident such as a fall or car crash. Following events like these, a

  3. #3 peak

    yourself, is it worth risking my health for the sake of “looking cooler?”⏎The risk of head injury might be low. But the risk of a car crash is also low, and yet you still wear a seatbelt. The risk of

  4. #4 peak

    or frequent pressure can irritate a bursa and cause inflammation.⏎Another cause of bursitis is a traumatic injury. Following trauma, such as a car accident or fall, a patient may develop bursitis. Usually a contusion causes

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. L1:535 suppresses
  2. L1:541 weakly excites
  3. L1:388 weakly suppresses
  4. L0:1571 weakly excites
  5. L1:1810 weakly suppresses
  6. L0:2782 weakly suppresses
  7. L1:254 weakly suppresses
  8. L0:2075 weakly excites
  9. L0:2250 weakly excites
  10. L1:170 weakly excites
  11. L1:241 weakly suppresses
  12. L1:861 weakly excites
  13. L1:201 weakly excites
  14. L1:1187 weakly suppresses
  15. L0:1617 weakly suppresses
  16. L1:1097 weakly excites
  17. L1:454 weakly excites
  18. L1:2433 weakly suppresses
  19. L1:936 weakly excites
  20. L0:1795 weakly excites
  21. L1:1285 weakly suppresses
  22. L0:2602 weakly excites
  23. L0:24 weakly suppresses
  24. L1:732 weakly suppresses

Feeds later layers

  1. L13:1717 suppresses
  2. L14:1197 excites
  3. L6:2804 suppresses
  4. L5:182 weakly excites
  5. L3:2418 weakly excites
  6. L7:719 weakly suppresses
  7. L3:1423 weakly excites
  8. L4:1647 weakly excites
  9. L9:2761 weakly excites
  10. L9:931 weakly suppresses
  11. L14:2400 weakly excites
  12. L10:2218 weakly excites
  13. L27:2783 weakly excites
  14. L14:2370 weakly excites
  15. L11:1387 weakly excites
  16. L13:1562 weakly excites unclear: words that set up a qualifier
  17. L10:849 weakly excites
  18. L16:268 weakly excites
  19. L23:1801 weakly suppresses
  20. L10:1947 weakly suppresses
  21. L25:309 weakly excites
  22. L20:524 weakly suppresses
  23. L4:2197 weakly suppresses
  24. L7:1108 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 -1.86 and -1.761 values between -1.66 and -1.561 values between -1.47 and -1.371 values between -1.37 and -1.271 values between -1.27 and -1.171 values between -1.17 and -1.083 values between -0.98 and -0.881 values between -0.88 and -0.792 values between -0.79 and -0.695 values between -0.69 and -0.591 values between -0.59 and -0.4910 values between -0.49 and -0.4018 values between -0.40 and -0.3069 values between -0.30 and -0.20325 values between -0.20 and -0.101258 values between -0.10 and -0.011664 values between -0.01 and 0.09394 values between 0.09 and 0.1988 values between 0.19 and 0.2921 values between 0.29 and 0.3918 values between 0.39 and 0.487 values between 0.48 and 0.585 values between 0.58 and 0.684 values between 0.68 and 0.772 values between 0.77 and 0.872 values between 0.87 and 0.973 values between 1.26 and 1.361 values between 1.36 and 1.461 values between 1.94 and 2.04offnegativepositiveoften

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. 骄傲
  2. 侧
  3. i
  4. потеря
  5. notation
  6. 自己的
  7. loses
  8. asting
  9. �
  10. of

Suppresses

  1. iores
  2. abit
  3. abis
  4. SURE
  5. (disposing
  6. kaar
  7. miễn
  8. upe
  9. quia
  10. mdp