Neuron Explorer

Qwen3-0.6B

line break mid-sentence in wrapped text

A neuron in layer 16 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 hard line break that was put into the middle of a sentence, where the text wraps onto a new line. It is about text that was split by a line width, not about a topic. Turning it up did not produce a clear theme.

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 low vitamin B12 levels in this population.3⏎Vitamin B12 is one of the most complex vitamins. The measurement of serum levels and⏎the significance of the results are much debated in the literature.⏎This article discusses testing

  2. #2 peak

    that likely will be too coarse to⏎locate with a GPS. You’ll need to find out where its located by⏎place name, some web research will likely be needed, though⏎sometimes the name

  3. #3 peak

    The atmosphere is⏎comprised of about 80% nitrogen, 16% oxygen, and⏎the rest is a mixture of other various gases.<|endoftext|>WASHINGTON — Scientists predicting the

  4. #4 peak

    projection.⏎15. Detail Drawing⏎Contains detail<|endoftext|>Rheumatic fever is an⏎inflammatory disease that may develop after an infection (strep throat or⏎scarlet fever) with group A Streptococcus

  5. #5 peak

    The 1992 fieldwork consisted of⏎surface collection and the excavation of five one-meter-square units. The⏎prehistoric portion of the site is badly disturbed by the historic⏎activities of

  6. #6 peak

    (called the Principal Act) which⏎received the assent of the President on the 15th April 1962 and was published in the⏎Official Gazette on the 2nd May 1962. The object

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: and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and,

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

    made.⏎But when they saw they were out-flanked,⏎They wheeled about and fled,⏎While many of our Northern braves⏎Were numbered with the dead⏎The double guide our soldiers took,⏎Soon left me far behind—⏎That aim

  2. #2 peak

    his maternal uncle or aunt will present them with a silk dhoti, and⏎the occasion will be celebrated as a sign of the boy's passage into maturity⏎and promising future. A significant event in the culture and history of South⏎India

  3. #3 peak

    Consequences, 1984, p.21).⏎Managerial performance can be defined as accomplishment of a given task measured against preset standards of accuracy, completeness, cost, and speed.⏎With the help of

  4. #4 peak

    a balance, both husband and wife enjoy their own well-defined rights in their own capacities, as quoted in the Holy Quran:⏎“And they (women) have rights (over their husbands) similar (to those of their husbands) over them to what

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. L14:191 excites blank spot before a run-on number
  2. L12:5 excites
  3. L14:960 excites
  4. L9:7 excites
  5. L14:125 suppresses
  6. L15:162 weakly suppresses
  7. L3:3 weakly suppresses
  8. L7:2996 weakly excites
  9. L10:346 weakly suppresses "each" in "each other" phrases
  10. L6:108 weakly suppresses
  11. L15:210 weakly excites
  12. L12:2015 weakly suppresses
  13. L10:184 weakly suppresses
  14. L8:101 weakly excites unclear: commas and 'or' in lists
  15. L6:1582 weakly excites
  16. L2:2897 weakly suppresses
  17. L8:30 weakly excites blank slot right before a number
  18. L12:372 weakly suppresses
  19. L14:51 weakly excites colon that introduces a statement
  20. L10:1224 weakly excites
  21. L6:0 weakly excites
  22. L14:211 weakly suppresses
  23. L0:2126 weakly suppresses
  24. L15:998 weakly excites period inside an abbreviation

Feeds later layers

  1. L17:146 strongly suppresses period after abbreviations and dates
  2. L19:1363 suppresses
  3. L17:585 excites
  4. L18:603 suppresses
  5. L21:81 suppresses
  6. L18:93 excites
  7. L17:354 excites
  8. L17:1772 excites
  9. L19:290 excites
  10. L19:413 suppresses
  11. L20:939 suppresses
  12. L17:694 suppresses
  13. L19:1506 excites
  14. L19:223 excites
  15. L27:262 excites
  16. L19:2302 suppresses
  17. L23:56 suppresses
  18. L25:52 excites
  19. L22:160 excites
  20. L18:313 excites
  21. L19:257 excites
  22. L19:255 excites
  23. L20:262 excites
  24. L19:335 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 -5.90 and -5.692 values between -5.27 and -5.051 values between -4.21 and -4.001 values between -3.58 and -3.371 values between -3.37 and -3.161 values between -3.16 and -2.951 values between -2.95 and -2.741 values between -2.74 and -2.533 values between -2.32 and -2.112 values between -2.11 and -1.905 values between -1.90 and -1.695 values between -1.69 and -1.488 values between -1.48 and -1.274 values between -1.27 and -1.0611 values between -1.06 and -0.8514 values between -0.85 and -0.6433 values between -0.64 and -0.43265 values between -0.43 and -0.221196 values between -0.22 and -0.011467 values between -0.01 and 0.20614 values between 0.20 and 0.41219 values between 0.41 and 0.6242 values between 0.62 and 0.838 values between 0.83 and 1.041 values between 1.04 and 1.251 values between 1.88 and 2.091 values between 2.30 and 2.51offnegativepositiveoften

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. 但也
  2. 她说
  3. ernet
  4. <boost
  5. seau
  6. 她的
  7. tığını
  8. hers
  9. Pil
  10. 她在

Suppresses

  1. throughout
  2. mdb
  3. �
  4. vest
  5. пут
  6. INU
  7. Kohana
  8. within
  9. governed
  10. pressed