Neuron Explorer

Qwen3-0.6B

bracketed unit conversions

A neuron in layer 23 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 opening bracket that comes right after a measurement, such as "-75.5 °F (" or "40 F (", where the same value is given again in another unit. Its negative side fires on digits in years and ages, which is a different pattern. Steering effects were unrelated to units, so the neuron's role in the output is unclear.

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

    on Athabaskan name meaning “The Great One”). The majestic mountain, which has extremely cold weather with temperatures as low as -75.5 °F (−60 °C) and wind chills as low as −8.

  2. #2 peak

    anna Climate).⏎The average temperature for the year in Mount Surprise is 73.0°F (22.8°C). The warmest month, on average, is November

  3. #3 peak

    bonds with a magnetic field via a device such as a laser. The record for the hottest plasma formed on Earth measures a scorching 6 trillion degrees Celsius (10.8 trillion °F)! Definitely hotter than the temperature of the centre

  4. #4 peak

    112 people, 46 households, and 32 families residing in the town. The population density was 1.2 people per square mile (0.4/km²). There were 106 housing units at

  5. #5 peak

    around the world are collaborating on solving the speed problem, and they’ve come up with a new technique that promises to one day record terabytes of data per second (even the fastest SSDs only manage around 500MB/s today).⏎

  6. #6 peak

    minated and externally developed school improvement programs known as "whole-school" or "comprehensive" reforms. In addition to reviewing the overall achievement effects of comprehensive school reform (CSR), they study the specific effects of 29 of the most widely implemented

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: The use of a certain kind of form of communication at some level might form an identity or maybe even a form of communication within certain form of communication. What I mean by that is that if you use form of communication techniques or form of communication

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

    Juno launched in 2011 and arrived at Jupiter in 2016. It has since completed 47 close flybys of the planet

  2. #2 peak

    sparklers are sometimes considered to be safe, they were responsible for 17 percent of the injuries, including 36 percent of those sustained by children under 5 years of age, according to the U.S. Consumer Product Safety Commission (

  3. #3 peak

    involved in 28,496 deaths, the study says. More than 8,400 of these occurred among adults between the ages of 25 and 34, a number high enough to mean that 2

  4. #4 peak

    deaths in this demographic.⏎While opioid-related deaths were common among older age groups as well — there were about 6,700 among adults ages 35 and 44, more than 5,600 among

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. L14:348 suppresses
  2. L9:7 excites
  3. L21:219 excites
  4. L5:5 suppresses digit after a label like paper or type
  5. L16:723 suppresses
  6. L21:173 excites
  7. L20:527 excites
  8. L7:6 suppresses
  9. L10:33 suppresses
  10. L12:940 excites
  11. L21:327 weakly excites
  12. L15:44 weakly suppresses
  13. L6:158 weakly excites
  14. L17:533 weakly excites
  15. L15:123 weakly suppresses
  16. L19:981 weakly excites
  17. L21:596 weakly suppresses
  18. L11:31 weakly suppresses
  19. L20:556 weakly excites
  20. L16:286 weakly excites
  21. L19:1697 weakly excites
  22. L20:959 weakly suppresses
  23. L18:48 weakly suppresses
  24. L22:1055 weakly suppresses

Feeds later layers

  1. L25:239 excites
  2. L26:17 excites
  3. L27:83 excites
  4. L27:344 excites
  5. L27:101 excites
  6. L27:318 excites opening parenthesis after a name or term
  7. L27:512 excites
  8. L25:222 excites
  9. L27:38 excites
  10. L26:263 excites
  11. L27:64 suppresses
  12. L27:431 excites
  13. L27:483 excites
  14. L27:513 excites
  15. L27:17 excites
  16. L27:558 excites
  17. L27:261 excites
  18. L25:265 excites
  19. L27:192 excites
  20. L26:261 excites
  21. L27:458 excites
  22. L27:632 excites
  23. L27:195 excites
  24. L25:15 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 -18.50 and -17.181 values between -15.86 and -14.543 values between -13.22 and -11.904 values between -11.90 and -10.5814 values between -10.58 and -9.2616 values between -9.26 and -7.9440 values between -7.94 and -6.6298 values between -6.62 and -5.30166 values between -5.30 and -3.98317 values between -3.98 and -2.66507 values between -2.66 and -1.341215 values between -1.34 and -0.021124 values between -0.02 and 1.31184 values between 1.31 and 2.6396 values between 2.63 and 3.9558 values between 3.95 and 5.2718 values between 5.27 and 6.5914 values between 6.59 and 7.918 values between 7.91 and 9.236 values between 9.23 and 10.555 values between 10.55 and 11.871 values between 11.87 and 13.191 values between 13.19 and 14.511 values between 15.83 and 17.153 values between 17.15 and 18.471 values between 19.79 and 21.111 values between 22.43 and 23.751 values between 23.75 and 25.071 values between 25.07 and 26.392 values between 26.39 and 27.711 values between 32.99 and 34.31offnegativepositiveoften

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. to
  2. den
  3. pre
  4. down
  5. table
  6. te
  7. nam
  8. it
  9. co
  10. ne

Suppresses

  1. @admin
  2. [Test
  3. @d
  4. @Test
  5. @g
  6. inux
  7. @m
  8. @register
  9. /Library
  10. ibraries