Neuron Explorer

Qwen3-0.6B

first piece of a split word

A neuron in layer 26 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

It reacts to the opening piece of a longer word that is broken into parts, like 'Suk' in Sukhothai, 'Yuk' in Yukon or 'techn' in technologically. When it was turned down, the model kept coming back to words like 'mind' and 'eye', and when turned up it repeated 'interesting'.

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

    populations throughout the region often referred to as the 'deep south'.⏎The old ethnic Malay and Muslim Sultanate of Patani made tributes to the ancient kingdoms of Sukhoth

  2. #2 peak

    9), and the Piemonte-Toscana region of northern/central Italy (images only, February-March, 2009).⏎In Fort Yukon, Alaska, salmon is a mainstay of the diet. Yet, among

  3. #3 peak

    them, and this comes from close observation of the plants plus a large application of common sense. Just take one step at a time. Master the fundamentals before you succumb to the excitement of special techniques or the challenge of difficult subjects

  4. #4 peak

    new methods to make their classes more effective, active, technologic and funny. Games and technologic tools (such as; web tools digital

  5. #5 peak

    land⏎(1808–1867) Hannah was born and raised in a Quaker family in Sherwood, NY and is best recognized for her work as an abolitionist. The

  6. #6 peak

    the universe.⏎She has just completed 47 interviews back to back (Len Irish, who took our cover photo of Collins, says she was one of the nicest people he ever photographed – “no fuss”), there’s a television crew waiting

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: thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought thought

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

    American democracy has changed A LOT over the centuries. Some people find this fact a little disconcerting, since we often take comfort in the notion that amidst the rapid social

  2. #2 peak

    behavorialist schools of psychology that reigned in his day, seeking a richer and more complex view of the mind than either provided. In a long, incisive essay, Brian Boyd celebrates the psychological acuity of Nabokov's

  3. #3 peak

    well as many other groups, highlights areas where some progress has been made, and lays out the paths toward significant solutions.⏎Some shocking statistics⏎These statistics provide some enlightening examples of what the report found — and the discrepancies are dramatic.⏎- African

  4. #4 peak

    . This makes it much harder for MRSA to develop a resistance. Ninja Polymers are nontoxic<|endoftext|>Watching an athlete choke can be one of the most difficult, and at the same time, engaging moments to witness. Like watching the

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. L20:2747 excites
  2. L23:1241 excites
  3. L8:1049 suppresses
  4. L9:150 excites
  5. L2:2154 weakly suppresses
  6. L16:379 weakly suppresses
  7. L16:34 weakly suppresses
  8. L21:2954 weakly excites
  9. L5:609 weakly suppresses
  10. L1:14 weakly suppresses
  11. L23:270 weakly suppresses
  12. L25:2812 weakly suppresses
  13. L14:1631 weakly excites
  14. L19:134 weakly excites
  15. L15:2331 weakly suppresses
  16. L22:2088 weakly excites
  17. L7:3038 weakly excites
  18. L24:1299 weakly excites the word 'value' in the sense of worth
  19. L6:2675 weakly excites
  20. L25:2754 weakly suppresses
  21. L2:1918 weakly suppresses
  22. L20:541 weakly suppresses
  23. L9:2789 weakly excites
  24. L24:428 weakly suppresses

Feeds later layers

  1. L27:1297 weakly suppresses
  2. L27:313 weakly suppresses
  3. L27:2988 weakly suppresses capital M starting a word
  4. L27:1436 weakly excites
  5. L27:1447 weakly excites
  6. L27:1090 weakly suppresses
  7. L27:2117 weakly excites
  8. L27:1099 weakly suppresses
  9. L27:202 weakly suppresses
  10. L27:361 weakly suppresses
  11. L27:486 weakly excites
  12. L27:415 weakly suppresses
  13. L27:1912 weakly suppresses
  14. L27:809 weakly suppresses
  15. L27:859 weakly excites unclear: token just before a k sound
  16. L27:2925 weakly excites
  17. L27:1538 weakly excites
  18. L27:1214 weakly excites
  19. L27:2753 weakly suppresses
  20. L27:2226 weakly suppresses
  21. L27:1832 weakly suppresses
  22. L27:888 weakly excites
  23. L27:2652 weakly suppresses
  24. L27:2825 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 -36.22 and -34.531 values between -26.06 and -24.371 values between -20.98 and -19.292 values between -17.60 and -15.901 values between -15.90 and -14.212 values between -14.21 and -12.523 values between -12.52 and -10.826 values between -10.82 and -9.138 values between -9.13 and -7.4412 values between -7.44 and -5.7514 values between -5.75 and -4.0539 values between -4.05 and -2.36371 values between -2.36 and -0.673200 values between -0.67 and 1.03145 values between 1.03 and 2.7260 values between 2.72 and 4.4116 values between 4.41 and 6.1112 values between 6.11 and 7.804 values between 7.80 and 9.492 values between 9.49 and 11.182 values between 11.18 and 12.881 values between 12.88 and 14.571 values between 14.57 and 16.261 values between 16.26 and 17.961 values between 19.65 and 21.341 values between 28.11 and 29.811 values between 29.81 and 31.50offnegativepositiveoften

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. 有趣
  2. eworthy
  3. utz
  4. (inter
  5. خاص
  6. interesting
  7. 有趣的
  8. interes
  9. Interesting
  10. df

Suppresses

  1. mind
  2. 眼
  3. 心
  4. Eye
  5. eye
  6. thought
  7. ⼼
  8. Eye
  9. 心灵
  10. heart