Neuron Explorer

Qwen3-0.6B

'since' after a change or growth

A neuron in layer 7 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 word 'since' when it comes after a sentence describing something that has changed or grown, like 'has improved since 2002' or 'nearly doubled since 2009'. It is one of the clearest patterns in this set. Steering the model with this neuron did not produce text that shows the same theme, though.

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

    newborn care. They then teach pregnant mothers and their families the simple but important actions they can take to preserve the lives of newborns.⏎Although general health has improved since 2002, Afghanistan remains a country with some of the poorest health

  2. #2 peak

    ⏎October 30, 2019⏎Uncle Sam needs you, so long as you are male.⏎Women’s equality has drastically increased since the passing of the 19th amendment 99 years ago: divorce

  3. #3 peak

    . This screening has continued with students in Years 2 and 3.⏎Boondall State School 's Year 3 Literacy Test results have increased significantly since the ELF program began. The percentage of students identified as being in the bottom

  4. #4 peak

    the phrase has shifted since then, referring not to patrons and their clients, but 'a collective political preference

  5. #5 peak

    sentences. The UN Working Group on Enforced or Involuntary Disappearances (WGEID) stated that while Chile has made great strides since it transitioned to democracy, many challenges remain. While the WGEID praised

  6. #6 peak

    involving both prescribed and illicit opioids are rising sharply. They’ve nearly doubled since 2009, and have infiltrated all genders, demographics and geographic

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: How to calculate the angles of a regular polygon (n) of a regular polygon? Remembering the information from the problem, but without the actual data. We can assume that the regular polygon is given in the form (n). Given a

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

    wax as lip balm.<|endoftext|>Health care employment has been the bright spot in the otherwise lackluster recent jobs reports. As overall employment decreased by 2 percent from 2000 to 2010, employment in the

  2. #2 peak

    waters provide excellent fishing grounds. Peru is the world's second largest producer of silver and copper.⏎The Peruvian economy grew by an average of 5.6% per year from 2009-13 with a stable exchange rate

  3. #3 peak

    to commercial use in the mid-1960s. Container use has continued to grow steadily in the United States and is projected to grow at 6 percent annually. Presently, 80 percent of U.S. overseas cargo trade

  4. #4 peak

    through various arts and crafts projects. Provide them with art<|endoftext|>Reverse transcriptase from Moloney murine leukemia virus (MMLVRT) is an RNA dependent DNA polymerase, which has been used as a fundamental tool for molecular biology and biotechnology

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. L6:2539 suppresses
  2. L5:2919 excites
  3. L6:1732 excites
  4. L6:1452 suppresses
  5. L5:685 suppresses
  6. L4:1290 excites
  7. L6:1890 suppresses
  8. L4:2220 weakly excites
  9. L4:1483 weakly excites
  10. L6:2052 weakly suppresses
  11. L1:2641 weakly suppresses
  12. L5:198 weakly excites unclear: weak "and" in lists
  13. L6:2979 weakly suppresses
  14. L6:801 weakly suppresses
  15. L3:2315 weakly suppresses
  16. L6:1282 weakly suppresses
  17. L4:605 weakly excites
  18. L3:2898 weakly suppresses
  19. L6:2665 weakly suppresses
  20. L6:2044 weakly suppresses
  21. L4:266 weakly excites
  22. L5:1691 weakly suppresses
  23. L1:3033 weakly suppresses
  24. L5:1462 weakly suppresses

Feeds later layers

  1. L14:867 excites
  2. L12:2924 excites
  3. L19:1499 excites
  4. L9:743 excites
  5. L11:1895 excites
  6. L9:2603 excites
  7. L17:1071 excites
  8. L11:239 suppresses
  9. L22:1555 excites
  10. L8:2196 excites comparison words like more, same, but
  11. L8:1732 excites
  12. L12:1177 excites
  13. L12:2918 excites
  14. L11:1755 excites
  15. L21:1734 excites
  16. L12:1122 excites
  17. L13:2961 excites
  18. L8:3049 excites
  19. L17:2739 suppresses
  20. L9:2310 weakly excites
  21. L9:803 weakly excites
  22. L23:477 weakly excites
  23. L10:2139 weakly excites
  24. L16:1678 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 -2.18 and -2.003 values between -1.45 and -1.272 values between -1.27 and -1.092 values between -1.09 and -0.914 values between -0.91 and -0.735 values between -0.73 and -0.5542 values between -0.55 and -0.36283 values between -0.36 and -0.181607 values between -0.18 and -0.001632 values between -0.00 and 0.18260 values between 0.18 and 0.3638 values between 0.36 and 0.548 values between 0.54 and 0.723 values between 0.72 and 0.903 values between 0.90 and 1.093 values between 1.09 and 1.274 values between 1.45 and 1.632 values between 1.63 and 1.811 values between 1.99 and 2.171 values between 2.17 and 2.351 values between 3.62 and 3.801 values between 3.80 and 3.981 values between 4.17 and 4.351 values between 4.89 and 5.07offnegativepositiveoften

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. differences
  2. changes
  3. _changes
  4. _changed
  5. 的变化
  6. difference
  7. 变化
  8. difference
  9. 改变
  10. 変わって

Suppresses

  1. ную
  2. айн
  3. очный
  4. 股市
  5. ydı
  6. GLISH
  7. ][/
  8. 科学院
  9. aviest
  10. administering