Neuron Explorer

Qwen3-0.6B

"to" introducing a purpose

A neuron in layer 16 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 "to" when it introduces why something is done, like "a technique to improve" or "used to expand". It goes quiet on "to" in other roles, such as "the first person to learn". Turning it up made the model's continuations wander without any 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

    artists include Katsushika Hokusai,<|endoftext|>Weight loss can be a major struggle for some people, so much so that they turn to gastric bypass surgery to help them achieve their weight loss

  2. #2 peak

    believe that women are as capable of serving in combat as men are. Thus, the only changes to the draft system in the last 40 years have been to decrease draft dodging.⏎Around 210,485 out of

  3. #3 peak

    related to this resource.⏎You have successfully created an account.⏎A verifications link was sent to your email at . Please check your email and click on the link to confirm your email address and fully activate your iCPALMS account. Please check

  4. #4 peak

    “Stop living in the past”⏎“Look forward, not back”⏎All said by people who seemingly think that experiences of trauma can be consciously resolved, if we just try hard enough not to think about them, as if it were a choice to

  5. #5 peak

    , stifling growth. Fertilizers and pesticides have minimal effect because they cannot penetrate the soil, and ultimately wash away. Aerating your lawn is a technique to improve permeability. It counteracts compacted soil by removing small plugs from the

  6. #6 peak

    ander is often used in cases where the upper arch isn't spreading as wide as it needs to, says Dr. Richeson, and such an appliance is utilized to expand that arch. The palatal expander is best used while a child is

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: the process of a change of a process is a process of change in the whole of the process. The process of change is changing. The process of change is changing. The process of change is changing. The process of change is changing.

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

    months when his Dutch owner, one of the most powerful men in the colony, beat<|endoftext|>Manufacturers - 2⏎and (

  2. #2 peak

    of the love of reading between you and your child.<|endoftext|>U.S. politics and policy updates⏎Sign up for myFT Daily Digest and become the first person to learn about U.S. politics and policy news.⏎The Biden administration stated that in

  3. #3 peak

    diagnosing secondary breast cancer⏎The standard disclaimer is very relevant to this blog.⏎Disclaimer: Please note- This blog is NOT medical advice. This blog is purely for information only. See your own doctor to discuss concerns and options relevant for you.

  4. #4 peak

    don’t know where to start, consider keeping Synodontis catfish in your home aquarium! These hardy African catfish are popular freshwater aquarium fish due to their beautiful and unique patterning. They’re also relatively easy to care for, making

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. L15:1176 excites
  2. L12:1799 suppresses
  3. L12:237 excites
  4. L14:473 suppresses
  5. L8:723 suppresses
  6. L12:2212 suppresses 'to' or 'for' opening a purpose phrase
  7. L5:2132 excites
  8. L12:1607 excites
  9. L15:2262 suppresses
  10. L13:909 excites 'to' before a verb after a noun
  11. L15:1234 weakly suppresses
  12. L15:1320 weakly excites
  13. L14:2340 weakly suppresses
  14. L14:1158 weakly excites 'to' starting a purpose clause
  15. L8:94 weakly excites
  16. L14:867 weakly suppresses
  17. L14:77 weakly excites
  18. L15:901 weakly suppresses
  19. L15:1593 weakly excites
  20. L15:1837 weakly excites
  21. L13:1447 weakly suppresses
  22. L15:2611 weakly excites
  23. L15:1836 weakly excites
  24. L12:1032 weakly excites

Feeds later layers

  1. L17:591 weakly excites
  2. L26:2421 weakly excites
  3. L18:2437 weakly suppresses
  4. L25:1156 weakly suppresses
  5. L20:735 weakly excites
  6. L17:2351 weakly excites
  7. L23:1587 weakly suppresses
  8. L23:1370 weakly suppresses
  9. L17:1969 weakly excites
  10. L22:882 weakly excites
  11. L18:2326 weakly suppresses
  12. L19:1220 weakly suppresses
  13. L20:14 weakly excites
  14. L20:1613 weakly suppresses
  15. L17:2383 weakly excites
  16. L22:2962 weakly suppresses
  17. L18:2551 weakly excites
  18. L17:3026 weakly suppresses space right before a number
  19. L23:297 weakly suppresses
  20. L19:1780 weakly suppresses
  21. L22:1241 weakly suppresses
  22. L17:2371 weakly suppresses
  23. L22:610 weakly suppresses
  24. L21:2742 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 -3.36 and -3.181 values between -3.00 and -2.821 values between -2.82 and -2.641 values between -2.64 and -2.464 values between -2.46 and -2.283 values between -2.28 and -2.103 values between -2.10 and -1.928 values between -1.92 and -1.749 values between -1.74 and -1.5615 values between -1.56 and -1.3824 values between -1.38 and -1.2041 values between -1.20 and -1.0258 values between -1.02 and -0.8492 values between -0.84 and -0.66150 values between -0.66 and -0.48250 values between -0.48 and -0.30441 values between -0.30 and -0.121752 values between -0.12 and 0.06627 values between 0.06 and 0.24202 values between 0.24 and 0.4279 values between 0.42 and 0.6045 values between 0.60 and 0.7824 values between 0.78 and 0.9623 values between 0.96 and 1.1417 values between 1.14 and 1.328 values between 1.32 and 1.506 values between 1.50 and 1.683 values between 1.68 and 1.864 values between 1.86 and 2.045 values between 2.04 and 2.221 values between 2.40 and 2.583 values between 2.76 and 2.941 values between 2.94 and 3.122 values between 3.12 and 3.301 values between 3.30 and 3.481 values between 3.48 and 3.662 values between 3.66 and 3.84offnegativepositiveoften

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. 因为它
  2. عي
  3. ք
  4. mol
  5. olid
  6. .Closed
  7. duy
  8. Languages
  9. kilometers
  10. icates

Suppresses

  1. tdown
  2. leur
  3. dwelling
  4. runnable
  5. vant
  6. presently
  7. usr
  8. ANDING
  9. userdata
  10. uite