Neuron Explorer

Qwen3-0.6B

the prefix 'con' in word fragments

A neuron in layer 26 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 fires on the syllable 'con' when it starts a word split into pieces, like 'con-notations' or 'con-cur'. Turning it up made the model repeat words like 'relative' and 'prioritization' over and over.

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

    fowl. Slang doesn’t usually have a euphemistic quality. When we think of euphemisms, we think of words that are substituted because their connotations are less distressing than the words they replace. In slang you frequently have

  2. #2 peak

    the Official Handbook of the Marvel Universe states that the character Northstar maintains speeds which can be considered "modest," especially when carrying a passenger. The Handbook also concedes that a solid object moving in the Earth’s atmosphere at several times the speed

  3. #3 peak

    a significant inequality exists, and the inequality indirectly or directly affects the greater good. Disparity and discrimination are components that aid in keeping racism alive. Researches strongly concur that minorities are overrepresented in all 3 components of the Criminal Justice System

  4. #4 peak

    a much more common site of endometriosis than the small bowel. Some symptoms are due to adhesions constricting, twisting, or pulling on the bowel.⏎Diagnosis relies a great

  5. #5 peak

    probably apply. Yet collaborative graphic design is not widespread in the Wiki world.⏎Image production vs. graphic design⏎It may be fair, in a sense, to conflate image production with graphic design. For example, contributors to WikiPedia often

  6. #6 peak

    centered and bathed in sunlight, has become covered in flour while he hurriedly scoops up every last grain from a looted depot. The photograph powerfully conveys the animalistic condition of the city's population in the aftermath of the conflict

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: prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior prior

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

    place on Earth.⏎Why Does Melting Ice Matter?⏎This rise in temperature dramatically throws off the balance of the Antarctic ecosystem.

  2. #2 peak

    ally, the factors that may limit uptake–in order to shed light on how to improve the promotion of food literacy skills. We will also hear from Erin Beagle, Executive Director of Roots to Harvest, who will share her insights and experiences addressing

  3. #3 peak

    Among these are your immune system and your metabolism. For the immune system, you may be temporarily weakened and prone to sickness, particularly if the weight gain or loss happened over a short period of time (as is the case with crash diets and extended

  4. #4 peak

    noting compared to the ongoing, major forces that shape planet Earth. In a geologic sense, there are greater forces at work, from the coming and going of ice ages to an ongoing imbalance in the planet's rotation that make it wobble like

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. L21:140 suppresses
  2. L19:724 excites
  3. L22:194 suppresses
  4. L17:264 excites
  5. L20:762 suppresses
  6. L13:132 suppresses "of" after a noun
  7. L16:1999 excites
  8. L19:1501 excites
  9. L24:424 excites
  10. L19:1539 excites
  11. L18:277 suppresses
  12. L22:145 suppresses
  13. L13:975 suppresses
  14. L15:2021 suppresses
  15. L20:519 weakly excites
  16. L19:2406 weakly suppresses
  17. L8:45 weakly excites
  18. L6:439 weakly suppresses
  19. L18:602 weakly excites
  20. L24:35 weakly suppresses
  21. L19:260 weakly excites
  22. L19:535 weakly suppresses
  23. L20:728 weakly excites
  24. L19:2863 weakly suppresses

Feeds later layers

  1. L27:1193 weakly suppresses
  2. L27:2444 weakly suppresses
  3. L27:2155 weakly suppresses
  4. L27:915 weakly excites unclear: mixed word fragments and phrases
  5. L27:1237 weakly suppresses
  6. L27:869 weakly suppresses
  7. L27:776 weakly excites
  8. L27:716 weakly suppresses
  9. L27:1361 weakly excites
  10. L27:2205 weakly suppresses
  11. L27:1566 weakly excites
  12. L27:718 weakly excites
  13. L27:2869 weakly excites
  14. L27:2854 weakly suppresses
  15. L27:661 weakly suppresses
  16. L27:2246 weakly suppresses the letter r starting a word fragment
  17. L27:1313 weakly suppresses
  18. L27:193 weakly excites
  19. L27:2949 weakly excites
  20. L27:885 weakly suppresses
  21. L27:2170 weakly excites
  22. L27:1030 weakly excites
  23. L27:1815 weakly suppresses
  24. L27:1208 weakly excites

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 -42.78 and -41.271 values between -38.26 and -36.751 values between -36.75 and -35.241 values between -33.73 and -32.233 values between -30.72 and -29.215 values between -29.21 and -27.706 values between -27.70 and -26.203 values between -26.20 and -24.695 values between -24.69 and -23.1811 values between -23.18 and -21.6713 values between -21.67 and -20.1619 values between -20.16 and -18.6625 values between -18.66 and -17.1539 values between -17.15 and -15.6450 values between -15.64 and -14.1367 values between -14.13 and -12.6393 values between -12.63 and -11.12104 values between -11.12 and -9.61123 values between -9.61 and -8.10161 values between -8.10 and -6.59198 values between -6.59 and -5.09219 values between -5.09 and -3.58253 values between -3.58 and -2.07424 values between -2.07 and -0.56637 values between -0.56 and 0.95525 values between 0.95 and 2.45377 values between 2.45 and 3.96226 values between 3.96 and 5.47147 values between 5.47 and 6.9871 values between 6.98 and 8.4846 values between 8.48 and 9.9922 values between 9.99 and 11.5010 values between 11.50 and 13.0115 values between 13.01 and 14.523 values between 14.52 and 16.024 values between 16.02 and 17.53offnegativepositiveoften

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. 也有一些
  2. 一回事
  3. RoleId
  4. ContextHolder
  5. OwnerId
  6. 在那里
  7. Setter
  8. ponse
  9. LayoutInflater
  10. setName

Suppresses

  1. starting
  2. prior
  3. near
  4. bearing
  5. relative
  6. near
  7. ess
  8. starting
  9. operating
  10. spoken