Neuron Explorer

Qwen3-0.6B

comma after a subject phrase

A neuron in layer 19 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 responds to a comma that comes right after a subject noun phrase and before a parenthetical like 'therefore' or 'also called'. The effect is weak compared with the other neurons here. Turning it down made the text loop on 'is there'.

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

    not have to be drawn. Successful root canal treatment can preserve the tooth for many years or even decades.⏎Root canal treatment: when needed?⏎A root canal treatment, also called root canal treatment, is often the last resort to save a sick or

  2. #2 peak

    treatment of rheumatism and gout. Comfrey has been used in folk medicine as a poultice for treating burns and wounds. However, internal consumption, such as in the form of herbal tea, is discouraged, as it has been

  3. #3 peak

    Concrete poured and formed on the construction site by concrete contractors is obviously at the mercy of weather conditions at the time of curing. Precast concrete panel construction, therefore, results in a stronger, less porous, higher quality structural component, with

  4. #4 peak

    -invasive nature. This is particularly true for COVID-19 cases, where keeping patients comfortable and stable is critical.⏎- No Radiation⏎CT and X-ray, the two modalities most commonly used for lung diagnostics related to COVID-19

  5. #5 peak

    air. Its operation is based on the ventilation unit with a heat exchanger, which is responsible for supplying cool air in summer and warm in winter⏎This solution, therefore, requires the use of appropriate installation in each room, as well

  6. #6 peak

    hours a week for science, technology, history, geography and creative arts; and two to three hours a week for sport and other physical activities.⏎English and maths, which dominate the proposed mandatory outcomes, should make up 45 to 5

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: is there is a can can is is is is is is is is is is is is is is is is is is is there a is is is is is is is is there is is is is is is is there is is is is

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

    as a comorbid disorder with ADHD, Inattentive Type results in slower processing speed than it does in kids with ADHD, Hyperactive or Combined Type who also have a diagnosis of Anxiety Disorder. These children may also be identified as

  2. #2 peak

    the knowledge of laws that appear to govern the life of the past, obtained through investigation and analysis of the past events.⏎The stuff with which the traditional history is concerned, i.e. the events and incidents of the past, provides the rudimentary

  3. #3 peak

    : "It is the custom of our Government officials to struggle to make reasons for not doing things they obviously ought to do. When the particular thing they ought to do involves the encouragement of art, their struggles to shelve the whole matter become quite

  4. #4 peak

    blue dye DAPI in the second column. PK-M2 is visualized using a red stain in the third column, with the merge of the images in each row in the forth column. The 2nd and

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. L15:304 strongly excites
  2. L16:731 strongly excites
  3. L18:843 strongly excites
  4. L15:824 suppresses
  5. L17:1356 excites
  6. L18:162 suppresses
  7. L16:1442 suppresses
  8. L16:930 suppresses
  9. L10:422 excites
  10. L14:168 suppresses
  11. L15:2003 suppresses
  12. L15:318 suppresses
  13. L12:616 excites
  14. L17:626 excites
  15. L18:313 suppresses
  16. L18:1154 excites
  17. L13:706 excites
  18. L13:1995 suppresses
  19. L17:975 excites
  20. L10:586 suppresses 'important' and degree words after 'how'
  21. L14:1487 excites
  22. L16:1166 suppresses
  23. L15:916 suppresses
  24. L16:1223 excites

Feeds later layers

  1. L20:813 strongly suppresses
  2. L26:318 suppresses
  3. L24:240 suppresses
  4. L22:228 suppresses
  5. L21:1074 suppresses
  6. L23:220 suppresses
  7. L20:578 suppresses
  8. L21:262 excites
  9. L20:586 suppresses
  10. L22:479 suppresses
  11. L20:2046 suppresses
  12. L25:9 excites
  13. L21:594 suppresses
  14. L20:554 suppresses
  15. L20:1106 suppresses
  16. L20:1762 suppresses
  17. L21:359 suppresses
  18. L24:35 excites
  19. L23:127 excites
  20. L27:604 suppresses
  21. L23:257 excites
  22. L20:349 excites
  23. L20:810 suppresses
  24. L21:299 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 -5.68 and -5.412 values between -5.41 and -5.131 values between -3.78 and -3.511 values between -3.51 and -3.241 values between -3.24 and -2.974 values between -2.97 and -2.703 values between -2.70 and -2.437 values between -2.43 and -2.165 values between -2.16 and -1.896 values between -1.89 and -1.6114 values between -1.61 and -1.3418 values between -1.34 and -1.0733 values between -1.07 and -0.8081 values between -0.80 and -0.53462 values between -0.53 and -0.261520 values between -0.26 and 0.011018 values between 0.01 and 0.28441 values between 0.28 and 0.55165 values between 0.55 and 0.8245 values between 0.82 and 1.0915 values between 1.09 and 1.3714 values between 1.37 and 1.6414 values between 1.64 and 1.9111 values between 1.91 and 2.187 values between 2.18 and 2.457 values between 2.45 and 2.725 values between 2.72 and 2.992 values between 2.99 and 3.261 values between 3.26 and 3.531 values between 3.53 and 3.802 values between 3.80 and 4.071 values between 4.89 and 5.16offnegativepositiveoften

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. 哪
  2. ENSE
  3. aton
  4. енно
  5. ensa
  6. 级别的
  7. LOT
  8. iteur
  9. 尤其是
  10. its

Suppresses

  1. could
  2. can
  3. becomes
  4. applies
  5. became
  6. があります
  7. comes
  8. になります
  9. exists
  10. 进行了