Neuron Explorer

Qwen3-0.6B

"of" in "on top of"

A neuron in layer 2 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 lights up on the word "of" when it comes right after "on top", as in "on top of the soil" or "on top of each other". Turning it up made the model's writing drift into rambling, unrelated text, so the steering did not show a clear theme.

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

    ontan came in a predictable shape.⏎One could easily stack a few briquettes on top of each other in a stove and leave them burning slowly for hours. In addition,

  2. #2 peak

    arsenide. “We now demonstrate why it isn’t going to work”, Ron Jansen says. “If you put a layer of magnetic material directly on top of silicon, a barrier will form and the resistance is a factor of 10

  3. #3 peak

    you’d like to see.⏎You can create inverted alpha mattes on top of objects to effectively cut holes into other objects if you were to place one on top of another. If you want to duplicate a layer or layers, highlight the ones you

  4. #4 peak

    starts. There are many such alarms that incorporate vibrating sensations or audible alarms triggered by wet-detection devices that are attached to the underwear or a special pad on top of the bedding. The concept of the alarm treatment uses classic conditioning technique.⏎Once the

  5. #5 peak

    mattress construction.⏎How Memory Foam Beds Are Made⏎Memory foam mattresses are made up of regular polyurethane as the base layer. The memory foam sits on top of this, along with any other padding layers the mattress may need. Manufacturers often use

  6. #6 peak

    your pile and then wait for it to break down. No, you have to wait for the pile to break down before adding more dog waste on top of it.⏎So bottom line is, the nutrients in worm castings are good<|endoftext|>

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: to the present, how to get the application? 24 students. the college course (CS. 4 � law you exam (1 to 1注册考试() f (

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

    on November 8 in the year of 1908 Raja Rao was a famous writer. The place was Hassan, in the state of Mysore in south India, into a well-known Brahman family. He post-grad

  2. #2 peak

    Hindu traces the phrase back to a seminal 1955 essay by the sociologist M N Srinivas titled 'The Social System of a Mysore Village', quoting

  3. #3 peak

    be said of cyberspace.⏎Reflective Essay On Cyber Ethics⏎Essay On Cyberspace - Words | Cram⏎Parents of students and teachers should be

  4. #4 peak

    9: Differentiate between www and internet.⏎problem 10: What do you mean by Usenet? List out main categories of it?⏎problem 11: Define an OS. Describe in

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. L1:770 excites
  2. L1:1354 excites
  3. L0:191 weakly suppresses
  4. L1:666 weakly suppresses
  5. L0:1124 weakly suppresses
  6. L0:1778 weakly suppresses
  7. L1:2276 weakly excites
  8. L1:189 weakly suppresses
  9. L1:2318 weakly suppresses
  10. L1:754 weakly excites
  11. L1:2615 weakly excites
  12. L0:217 weakly suppresses
  13. L1:1208 weakly suppresses
  14. L1:1591 weakly excites
  15. L0:2455 weakly excites
  16. L1:2140 weakly excites
  17. L1:3003 weakly excites
  18. L1:1341 weakly excites
  19. L1:2574 weakly suppresses
  20. L0:1558 weakly excites
  21. L0:1809 weakly suppresses
  22. L1:1979 weakly suppresses
  23. L0:504 weakly excites
  24. L1:53 weakly excites

Feeds later layers

  1. L7:546 strongly suppresses
  2. L10:190 suppresses
  3. L3:318 excites
  4. L8:495 suppresses
  5. L9:144 suppresses
  6. L9:2483 suppresses
  7. L23:397 suppresses
  8. L16:692 suppresses
  9. L9:1625 suppresses
  10. L17:664 suppresses
  11. L19:1228 suppresses
  12. L20:500 suppresses
  13. L9:812 suppresses
  14. L10:1154 suppresses
  15. L24:2021 suppresses
  16. L24:2292 suppresses
  17. L18:177 excites
  18. L26:1783 suppresses
  19. L24:2599 weakly suppresses
  20. L11:2985 weakly suppresses
  21. L6:1504 weakly suppresses
  22. L24:2278 weakly suppresses 'Indian' before Ocean or subcontinent
  23. L21:901 weakly suppresses
  24. L9:453 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 -1.64 and -1.561 values between -1.15 and -1.061 values between -0.98 and -0.902 values between -0.73 and -0.652 values between -0.57 and -0.483 values between -0.48 and -0.4019 values between -0.40 and -0.3265 values between -0.32 and -0.24206 values between -0.24 and -0.15506 values between -0.15 and -0.071430 values between -0.07 and 0.011046 values between 0.01 and 0.10440 values between 0.10 and 0.18116 values between 0.18 and 0.2622 values between 0.26 and 0.3421 values between 0.34 and 0.435 values between 0.43 and 0.513 values between 0.51 and 0.593 values between 0.59 and 0.684 values between 0.68 and 0.761 values between 0.76 and 0.842 values between 0.84 and 0.922 values between 0.92 and 1.012 values between 1.01 and 1.091 values between 1.09 and 1.171 values between 1.17 and 1.251 values between 1.25 and 1.341 values between 1.42 and 1.501 values between 1.59 and 1.67offnegativepositiveoften

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. dT
  3. forgetting
  4. {}
  5. .UserService
  6. dd
  7. Bel
  8. name
  9. d
  10. Name

Suppresses

  1. lakh
  2. crore
  3. 男子
  4. haar
  5. Rajasthan
  6. antha
  7. ennai
  8. Gujar
  9. Jain
  10. arna