Neuron Explorer

Qwen3-0.6B

comma closing an introductory clause

A neuron in layer 11 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 a comma that ends an opening phrase such as 'If so,' or 'When you are interviewing contractors,', just before the main clause starts. It is weaker and less consistent on the negative side. Turning it up did not produce any clear change in topic or style.

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

    could discuss thermal expansion valves and compressors, but I think you now have enough info to be "dangerous"!⏎When you are interviewing potential hvac contractors, you will hear phrases such as total cooling capacity, sensible and latent capacity, bt

  2. #2 peak

    don’t have academic role models who look like them.⏎When the U.S. Department of Education’s Office for Civil Rights mandated diversity trainings for this school district, everyone pulled together to make some serious changes for the better.⏎A weekly sampling of

  3. #3 peak

    family ranch. Before running for president, Zelaya’s priorities in politics were mainly decentralizing government and protecting forestry against foreign concessions. If history tells us anything, his turn toward a more populist brand of politics has more to do with the energy

  4. #4 peak

    technologies are mostly used in the off-grid solar industry, and are both good options when it comes to charging your battery efficiently. When differentiating PWM and MPPT, the discussion is not going to be about which one is better than the other.⏎

  5. #5 peak

    location on the genome, they could tell whether the disease-causing traits are increasing or decreasing over time.⏎"We asked the question: When we look at human evolution, which is still under way, are we evolving towards, or away, from disease

  6. #6 peak

    are affecting data quality? If so, what sources?⏎- What makes a duplicate a duplicate?⏎- Should records be merged

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 myself. I have been thinking about myself. I have been thinking about myself. I have been thinking about myself. I have been thinking about myself. I have been thinking about myself. I have been thinking about myself

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

    containing the sanctuary, Its halls and ante-chambers. Most of the frontal courts and pylons are in ruin but because of the execution of the reliefs, a visit is immensely worthwhile. For example, just beyond the eight remaining columns of

  2. #2 peak

    peaked?⏎This in some degree corresponds with Captain Cook's record of the irregularity of his compass when he passed near this part of the coast, in consequence of which he called the peaked island to the westward of the cape, Magnetical Island

  3. #3 peak

    carbazepine⏎Some medicines are not suitable for people with certain conditions, and sometimes a medicine may only be used if extra care is taken. For these reasons, before you start taking oxcarbazepine it is important that your doctor

  4. #4 peak

    part because of what George Washington and his soldiers achieved against all the odds that nature and a vastly superior military force could pit against them. … Because of their accomplishments, the waning days of 1776 were not the end of everything

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. L9:1579 suppresses
  2. L8:410 suppresses
  3. L6:2843 suppresses
  4. L8:3062 weakly excites
  5. L10:205 weakly excites
  6. L6:982 weakly excites
  7. L3:2785 weakly suppresses
  8. L7:2943 weakly suppresses
  9. L10:464 weakly suppresses
  10. L4:478 weakly suppresses
  11. L8:485 weakly suppresses
  12. L6:2820 weakly suppresses
  13. L8:170 weakly excites
  14. L3:771 weakly suppresses
  15. L5:2333 weakly suppresses
  16. L8:1195 weakly excites
  17. L7:884 weakly excites
  18. L1:305 weakly excites
  19. L6:2233 weakly excites
  20. L8:190 weakly excites
  21. L7:1137 weakly suppresses
  22. L9:417 weakly suppresses
  23. L4:1028 weakly excites
  24. L9:1050 weakly excites

Feeds later layers

  1. L13:2366 suppresses
  2. L15:2938 suppresses comma after an if-clause
  3. L12:1642 suppresses
  4. L14:957 excites
  5. L17:2228 excites
  6. L18:1537 suppresses
  7. L13:1375 suppresses
  8. L15:2521 suppresses
  9. L14:1853 weakly suppresses
  10. L12:1137 weakly excites
  11. L18:1500 weakly suppresses
  12. L16:2762 weakly suppresses
  13. L15:582 weakly excites
  14. L12:2704 weakly suppresses
  15. L12:2151 weakly suppresses
  16. L20:1370 weakly excites
  17. L14:2745 weakly excites
  18. L17:2496 weakly suppresses
  19. L14:709 weakly excites
  20. L12:2685 weakly suppresses
  21. L13:2669 weakly suppresses
  22. L14:2944 weakly suppresses
  23. L15:1427 weakly suppresses
  24. L27:3007 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.

2 values between -3.75 and -3.601 values between -2.57 and -2.421 values between -2.28 and -2.132 values between -2.13 and -1.981 values between -1.54 and -1.391 values between -1.39 and -1.256 values between -1.25 and -1.107 values between -1.10 and -0.957 values between -0.95 and -0.815 values between -0.81 and -0.6622 values between -0.66 and -0.5178 values between -0.51 and -0.37283 values between -0.37 and -0.22882 values between -0.22 and -0.071963 values between -0.07 and 0.08498 values between 0.08 and 0.2293 values between 0.22 and 0.3734 values between 0.37 and 0.5212 values between 0.52 and 0.674 values between 0.67 and 0.812 values between 0.81 and 0.961 values between 1.11 and 1.251 values between 1.25 and 1.401 values between 1.55 and 1.691 values between 1.99 and 2.14offnegativepositiveoften

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. HN
  2. NSNotification
  3. zro
  4. ##⏎⏎
  5. frm
  6. Story
  7. .son
  8. (ptr
  9. familial
  10. istros

Suppresses

  1. альн
  2. ляет
  3. 我们必须
  4. 重回
  5. ��
  6. alis
  7. ạch
  8. озд
  9. 丽
  10. .Master