Neuron Explorer

Qwen3-0.6B

text cut off at a document break

A neuron in layer 14 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

The neuron tends to light up on the last word or two before a document is cut off, right before a separator marker. The examples are mostly mid-sentence endings in school and health articles. Turning it up did not produce a clear theme in the writing.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    from pollen and insects, and carbohydrates come from the sugars found in nectar. They spend a large amount of time foraging for food, which indicates that carbohydrates are extremely important to them. Their diet is high in energy, and low in protein

  2. #2 peak

    offer them everything they need to be successful.⏎3. Preparing for Standardized Testing.⏎With how important performance on standardized tests is for students, teachers, schools<|endoftext|>Do you know your credit score? Despite the fact that it’s one of the

  3. #3 peak

    idea don’t use exclamation marks and italics. Remember to use your best writing voice when writing argumentative essays. You’ll often be asked to write<|endoftext|>by Mike Adams⏎I

  4. #4 peak

    their . place. So, the students, teachers and other professionals finally had to resume their studies and jobs in government institutions like schools, colleges, courts etc.<|endoftext|>I hope you are all well and that you enjoyed some of your weekend in the

  5. #5 peak

    student handbooks, and course syllabi to establish expectations and reduce confrontations between instructors and students.⏎Sourced through Scoop.it from: www.educause.edu<|endoftext|>NEW DELHI, India — Next to a railway line in Okhla,

  6. #6 peak

    is fun. You are more likely to keep with it if you choose something you like. A lot of people find it's more fun to exercise with someone else<|endoftext|>Poor sleep quality is significantly associated with mood disturbance and lower quality of life among obese

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 been thinking about lately, and I have been thinking about lately, and I'm thinking about lately, and I'm thinking about lately, and I'm thinking about lately. I'm thinking about lately, and I'm thinking about lately

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

    .⏎15. Detail Drawing⏎Contains detail<|endoftext|>Rheumatic fever is an⏎inflammatory disease that may develop after an infection (strep throat or⏎scarlet fever) with group A Streptococcus

  2. #2 peak

    publication in Physical Review Letters.⏎Gravitational waves are ripples in the universe caused by some of the most energetic cosmic events, from exploding stars to supermassive black hole mergers. As they propagate through space and time, gravitational waves cause

  3. #3 peak

    Earth from their unique point of view aboard the International Space Station, providing researchers with key data and recording changes over time from human-caused changes like urban growth and reservoir construction, to natural dynamic events such as hurricanes, floods and volcanic eruptions.⏎

  4. #4 peak

    index bread has a value of 70 and above. Examples of breads that are low in simple carbohydrate include multi grain, whole grain, rye, sour dough, soy, pumpernickel and pita. Breads that are

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. L13:258 suppresses
  2. L13:164 suppresses
  3. L12:114 suppresses
  4. L13:632 excites
  5. L9:203 suppresses
  6. L12:263 excites
  7. L12:121 excites
  8. L11:358 suppresses
  9. L13:489 excites
  10. L12:149 suppresses
  11. L10:2130 excites
  12. L6:13 suppresses
  13. L12:1359 excites
  14. L13:1488 suppresses
  15. L12:192 suppresses
  16. L8:10 suppresses
  17. L12:562 suppresses
  18. L13:1045 suppresses
  19. L8:4 suppresses
  20. L12:89 suppresses
  21. L13:999 excites
  22. L11:1139 suppresses
  23. L13:1787 suppresses
  24. L8:7 suppresses

Feeds later layers

  1. L16:1116 excites
  2. L15:1049 excites
  3. L15:597 excites
  4. L19:396 weakly suppresses 'just' meaning only
  5. L20:1159 weakly suppresses
  6. L15:805 weakly excites
  7. L22:589 weakly suppresses
  8. L25:237 weakly suppresses
  9. L21:1202 weakly suppresses
  10. L21:133 weakly suppresses
  11. L17:660 weakly suppresses
  12. L19:323 weakly suppresses
  13. L24:142 weakly suppresses
  14. L26:1002 weakly suppresses
  15. L21:350 weakly suppresses
  16. L16:2115 weakly suppresses
  17. L17:590 weakly excites
  18. L16:1427 weakly excites
  19. L22:994 weakly suppresses
  20. L22:1276 weakly suppresses
  21. L16:1092 weakly suppresses
  22. L26:1724 weakly suppresses
  23. L15:1071 weakly suppresses
  24. L15:1444 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 -2.25 and -2.112 values between -2.11 and -1.982 values between -1.98 and -1.847 values between -1.70 and -1.576 values between -1.57 and -1.435 values between -1.43 and -1.2913 values between -1.29 and -1.1626 values between -1.16 and -1.0228 values between -1.02 and -0.8838 values between -0.88 and -0.7560 values between -0.75 and -0.61101 values between -0.61 and -0.47149 values between -0.47 and -0.34185 values between -0.34 and -0.20322 values between -0.20 and -0.06492 values between -0.06 and 0.08514 values between 0.08 and 0.21430 values between 0.21 and 0.35379 values between 0.35 and 0.48322 values between 0.48 and 0.62249 values between 0.62 and 0.76192 values between 0.76 and 0.89135 values between 0.89 and 1.0395 values between 1.03 and 1.1755 values between 1.17 and 1.3034 values between 1.30 and 1.4424 values between 1.44 and 1.5812 values between 1.58 and 1.7111 values between 1.71 and 1.858 values between 1.85 and 1.992 values between 1.99 and 2.124 values between 2.12 and 2.262 values between 2.26 and 2.401 values between 2.53 and 2.671 values between 2.67 and 2.811 values between 3.08 and 3.22offnegativepositiveoften

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. Girlfriend
  2. 牠
  3. shall
  4. _fence
  5. スター
  6. illing
  7. Shall
  8. flate
  9. Photographer
  10. 듯

Suppresses

  1. äd
  2. :_
  3. 亚
  4. ep
  5. Ep
  6. :%
  7. aes
  8. mess
  9. مق
  10. mid