Neuron Explorer

Qwen3-0.6B

word after an introductory phrase

A neuron in layer 16 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron tends to fire on the word or comma that comes right after a short lead-in phrase, like 'In such cases,' or 'For this reason'. It goes quiet on modal words such as 'can' or 'should' after a hopeful or evaluative clause. Turning it up made the model repeat itself and drift.

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

    .⏎The command for unzipping the .tgz file using the .tar utility:⏎tar -xvzf <File name>.tgz⏎In this command -xvzf flags are used, and these flags pass commands to the

  2. #2 peak

    benefited from this Ritchie Dental Group blog article and look forward to your next visit.<|endoftext|>Sometimes it makes sense for a function to return multiple things. In such cases, it's common to just return them in a tuple: "return (count

  3. #3 peak

    are designed to enhance understanding andor to provide additional practice. Each rule or line in an accesslist provides a condition, either permit or deny. In computer networking, port numbers are part of the addressing information used to identify the senders and receivers

  4. #4 peak

    . In many cases, both events are similarly included. For example, in a publication created by a team of contributors, each contributor has the lawful right to sell his/her job under a copyright task contract

  5. #5 peak

    format applies to every source type⏎Reproduced from https://style.mla.org/whats-new/⏎Author. When a source has three or more authors, only the first one shown in the source is normally given

  6. #6 peak

    cover the period from the French colonial era of the seventeenth century to the creation of Nunavut in 1999.⏎The map descriptions provided below have been reproduced from the original maps.⏎First successful French settlements in North America: Port

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: if we look at the average of the numbers that are at least the average of the numbers that are a minimum to the average of the numbers that are a minimum to the average of the numbers that are a minimum, etc. So, the

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

    Israel Institute of Technology and the study’s principal investigator, said in a press release.⏎“The good news is that since the ability is there, we hope future treatments can be developed to activate it and employ the body’s own tools to cure these

  2. #2 peak

    useful tools for teachers and school leaders?⏎Teacher Evaluation 2.0 tackles this key question. It proposes six design standards that any rigorous and fair teacher evaluation system should meet. It offers a blueprint for better evaluations that can help every teacher succeed in

  3. #3 peak

    the sound of croaking Gartraps looking for mates. Each mated female Gartrap lays two eggs and defends them with her life. Very few creatures will even try to eat a Gartrap’s eggs. They can live for up to

  4. #4 peak

    fat chart are based on the biggest serving size available.⏎- These nutrition facts came directly from the USDA or manufacturer/restaurant.⏎- If you're using a calorie counter, remember that Fat, Carbs, and Protein calories are just close estimations

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. L15:284 suppresses
  2. L15:862 excites
  3. L14:96 excites
  4. L13:1625 excites
  5. L14:699 suppresses
  6. L12:2825 excites
  7. L14:1574 excites
  8. L14:2863 suppresses
  9. L15:1478 suppresses
  10. L15:2564 suppresses
  11. L15:2684 suppresses
  12. L14:855 excites
  13. L15:1098 suppresses
  14. L14:2247 suppresses
  15. L14:2381 excites
  16. L15:1261 excites
  17. L1:646 suppresses
  18. L7:2262 weakly suppresses
  19. L14:2290 weakly suppresses
  20. L12:1466 weakly suppresses
  21. L14:2176 weakly excites
  22. L15:176 weakly suppresses
  23. L12:2026 weakly suppresses
  24. L13:1212 weakly suppresses

Feeds later layers

  1. L17:2220 suppresses
  2. L20:1156 suppresses
  3. L17:1886 suppresses
  4. L18:2880 suppresses
  5. L22:1922 weakly suppresses
  6. L21:2248 weakly suppresses
  7. L18:1751 weakly suppresses
  8. L19:1285 weakly suppresses
  9. L17:2672 weakly suppresses
  10. L22:2214 weakly suppresses
  11. L25:978 weakly suppresses
  12. L24:1262 weakly suppresses
  13. L17:2451 weakly suppresses
  14. L23:1156 weakly suppresses
  15. L21:2403 weakly suppresses
  16. L22:186 weakly suppresses
  17. L21:1183 weakly suppresses
  18. L18:1446 weakly suppresses
  19. L21:221 weakly suppresses
  20. L17:2106 weakly suppresses 'is' in a definition sentence
  21. L22:2851 weakly suppresses
  22. L18:371 weakly suppresses
  23. L18:529 weakly suppresses
  24. L18:1194 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 -2.93 and -2.791 values between -2.51 and -2.361 values between -2.36 and -2.221 values between -2.08 and -1.941 values between -1.94 and -1.792 values between -1.79 and -1.652 values between -1.65 and -1.512 values between -1.51 and -1.372 values between -1.37 and -1.225 values between -1.22 and -1.083 values between -1.08 and -0.9411 values between -0.94 and -0.8015 values between -0.80 and -0.6516 values between -0.65 and -0.5142 values between -0.51 and -0.37154 values between -0.37 and -0.23496 values between -0.23 and -0.081688 values between -0.08 and 0.06628 values between 0.06 and 0.20315 values between 0.20 and 0.35158 values between 0.35 and 0.49119 values between 0.49 and 0.6374 values between 0.63 and 0.7752 values between 0.77 and 0.9231 values between 0.92 and 1.0634 values between 1.06 and 1.2023 values between 1.20 and 1.3410 values between 1.34 and 1.496 values between 1.49 and 1.635 values between 1.63 and 1.773 values between 1.77 and 1.914 values between 1.91 and 2.062 values between 2.20 and 2.341 values between 2.63 and 2.77offnegativepositiveoften

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. acity
  2. プログ
  3. ric
  4. ublic
  5. рост
  6. לבד
  7. arter
  8. argins
  9. Expansion
  10. WB

Suppresses

  1. referencing
  2. 借鉴
  3. ��
  4. iona
  5. 参照
  6. .Equal
  7. %"><
  8. =$
  9. 趋
  10. adhere