Neuron Explorer

Qwen3-0.6B

commas after citation numbers

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 the comma or period that comes right after a reference number or inside a citation, such as '1,2,3' or 'N.p., n.d.'. It is a punctuation detector for bibliographies and footnotes. Turning it up made the text stutter into repeated words, so the steering result tells us little.

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

    -nee-oh-sen-TEE-sis) or chorionic (pronounced kawr-ee-ON-ik) villus sampling.1,2,3⏎- Amniocentesis. This involves inserting a needle

  2. #2 peak

    treatment due to its antibacterial and anti-inflammatory properties. Other medicinal uses include using wild chervil to treat cancer, kidney stones, and cystitis.1,2⏎Cultivation and harvest:⏎Wild chervil has become naturalized

  3. #3 peak

    procedures; however, all methods of declaw were proven painful. Pain levels were the same six days into recovery regardless of procedure type in one study.5,6 Another study comparing pain from scalpel declaw vs. laser declaw showed

  4. #4 peak

    fluid or blood administration; however, an acute coagulopathy can also occur in severely injured patients independent of, or in addition to, these factors [2,3]. Several terms are used in the literature to refer to this condition, including

  5. #5 peak

    Users, 2012-2017). According to Foreign Policy, The

  6. #6 peak

    an uplifting feeling and, without doubt, a gift.<|endoftext|>June 8, 2000⏎Board of Education of⏎Ladies and Gentlemen,⏎Few are presented with an opportunity to change the course of history. Washington and Lincoln

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 and I, I. I and I, I, and I I I I I I I I I I I , I, and I, and I, now, I, all and I, now, I,

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

    change, despite their traditionally low-carbon lifestyles. The Karuk, in particular have experienced dramatic environmental changes in their forestlands and fisheries as a result

  2. #2 peak

    should also meet healthcare standards for washability, mold and microbial resistance, and fire safety. Ceiling panels with these characteristics are ideal

  3. #3 peak

    :⏎Excellent Thermal Conductivity of Copper:⏎Copper possesses outstanding thermal conductivity, surpassing stainless steel by 24 times. This is particularly crucial in distillation, where even heating of the mash ensures more precise temperature control, preventing the loss of

  4. #4 peak

    may also have been influenced by analysis limitations because of sample size, other sources of individual variation, or environmental conditions.⏎You have requested a machine translation

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. L9:77 excites commas in citations and name lists
  2. L10:335 excites commas in catalog, date, and list metadata
  3. L4:3 excites commas after citation fields
  4. L5:16 excites
  5. L2:66 excites
  6. L3:71 suppresses
  7. L1:86 weakly excites
  8. L6:620 weakly suppresses "and" in lists of three or more
  9. L5:91 weakly excites
  10. L8:1718 weakly suppresses
  11. L5:158 weakly suppresses
  12. L5:1823 weakly excites
  13. L0:1224 weakly suppresses
  14. L3:1089 weakly suppresses
  15. L2:1790 weakly suppresses
  16. L9:1580 weakly excites
  17. L7:1779 weakly suppresses
  18. L9:2514 weakly suppresses
  19. L0:2288 weakly suppresses
  20. L1:1879 weakly suppresses
  21. L7:1241 weakly excites
  22. L1:1404 weakly suppresses
  23. L0:178 weakly suppresses
  24. L6:1589 weakly excites

Feeds later layers

  1. L14:35 strongly suppresses
  2. L16:240 strongly suppresses comma after 'sometimes' or 'the truth'
  3. L18:366 suppresses comma inside numbers and dates
  4. L16:846 suppresses comma opening an aside or explanation
  5. L17:118 suppresses comma after a clause ending
  6. L15:588 suppresses comma after an introductory phrase
  7. L23:184 suppresses
  8. L19:726 suppresses comma inside large numbers
  9. L16:1082 suppresses comma linking two clauses after 'and'
  10. L20:476 suppresses commas that lead into a continuation
  11. L22:144 suppresses
  12. L19:1826 suppresses
  13. L18:2043 suppresses
  14. L20:1298 suppresses
  15. L17:2001 suppresses
  16. L19:2645 suppresses commas in the middle of a sentence
  17. L17:585 suppresses
  18. L16:241 excites
  19. L19:383 suppresses commas after a clause or list item
  20. L12:1557 suppresses comma after an introductory phrase
  21. L15:1370 suppresses
  22. L18:1394 suppresses
  23. L17:627 suppresses
  24. L17:116 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 -4.03 and -3.882 values between -3.74 and -3.592 values between -3.44 and -3.291 values between -2.99 and -2.854 values between -2.85 and -2.704 values between -2.70 and -2.552 values between -2.55 and -2.406 values between -2.40 and -2.251 values between -2.25 and -2.113 values between -2.11 and -1.964 values between -1.96 and -1.815 values between -1.81 and -1.664 values between -1.66 and -1.5110 values between -1.51 and -1.379 values between -1.37 and -1.2211 values between -1.22 and -1.0714 values between -1.07 and -0.926 values between -0.92 and -0.7717 values between -0.77 and -0.6215 values between -0.62 and -0.4823 values between -0.48 and -0.3368 values between -0.33 and -0.18359 values between -0.18 and -0.031295 values between -0.03 and 0.121272 values between 0.12 and 0.26586 values between 0.26 and 0.41151 values between 0.41 and 0.5625 values between 0.56 and 0.714 values between 0.71 and 0.862 values between 0.86 and 1.011 values between 1.01 and 1.151 values between 1.75 and 1.89offnegativepositiveoften

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. ,
  2. .setdefault
  3. 实在
  4. ,在
  5. fr
  6. iders
  7. าะ
  8. ,...
  9. 到位
  10. srv

Suppresses

  1. >equals
  2. aturas
  3. LESS
  4. tes
  5. buckle
  6. itin
  7. noon
  8. oulos
  9. ateau
  10. который