Neuron Explorer

Qwen3-0.6B

"Al-Aqsa" and "Lunar Module" name parts

A neuron in layer 5 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron responds to the second half of certain names, such as the "sa" in Al-Aqsa or "Module" in Lunar Module. It mostly stays quiet, and it strongly switches off on words like "ring" and "of" in "Kingdom of". Turning it up or down gave text that did not clearly connect to this.

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

    recalled that “248 took refuge there. They included 1 Islamic Jihad, 28 Hamas, 50 to 60 Al-Aqsa Martyrs. The remainder were ordinary townsfolk and included 100 uniform

  2. #2 peak

    the Israeli Supreme Court over the eviction of six Palestinian families from Sheikh Jarrah in occupied East Jerusalem.⏎- The next day, Israeli Police stormed the Al Aqsa mosque.⏎- A few days later, Hamas and other Palestinian groups started firing rockets

  3. #3 peak

    over the Pacific from an observation plane.⏎The European Space Agency's first cargo mission to the international space station ended in a spectacular fireworks show today, with the fiery re-entry of the

  4. #4 peak

    limestone and cullet⏎Information provided by the Glass Packaging Institute http://www.gpi.org.⏎Answers: C, A, B, B, C<|endoftext|>This article

  5. #5 peak

    third-stage boosters that were used for seismic studies and six Lunar Module Ascent Stages that were crashed at the end of their missions, as was

  6. #6 peak

    Pacific Ocean was colder than normal, but it was warmer than usual in the Atlantic. That combination caused a drought in Colorado that was just as bad as the Dust Bowl of the 1930s.⏎The ocean currents slipped back into their

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 set of all integers. The set of all integers is closed under addition and multiplication. Let's say, $ \mathbb{Z} $, and $ \mathbb{Z} $. Let's say $ \mathbb{

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

    and the operation of convolution define a commutative ring. Any ring without divisors of zero can be extended to a quotient field or

  2. #2 peak

    ai and Ayutthaya, and was incorporated into the Kingdom of Siam in 1786. Patani was divided into the three

  3. #3 peak

    chlorine/chloramine also reduces TTHM’s, insecticides, pesticides, synthetic fertilizers<|endoftext|>The Red Sea is a prime environment and resource for the Kingdom of Saudi Arabia. Centuries of shipping and fishing activities supported local communities. In recent

  4. #4 peak

    classically Diophantine geometry), a subfield of algebraic number theory.⏎- The study of the real points of an algebra

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. L4:2174 suppresses
  2. L4:2211 weakly excites
  3. L3:2845 weakly suppresses
  4. L4:829 weakly suppresses
  5. L3:1989 weakly suppresses
  6. L4:993 weakly excites
  7. L3:1042 weakly excites
  8. L4:1100 weakly suppresses
  9. L3:1858 weakly suppresses
  10. L4:385 weakly excites multiplication and scaling in math
  11. L3:1664 weakly suppresses acronym endings like HSUS and SDS
  12. L4:970 weakly excites
  13. L0:2296 weakly suppresses
  14. L1:1369 weakly suppresses
  15. L4:2281 weakly excites
  16. L0:1350 weakly suppresses
  17. L3:503 weakly excites
  18. L3:1715 weakly suppresses
  19. L4:2388 weakly suppresses
  20. L4:2598 weakly suppresses
  21. L4:2953 weakly excites
  22. L3:1727 weakly excites
  23. L2:336 weakly suppresses
  24. L2:2015 weakly suppresses

Feeds later layers

  1. L6:511 suppresses
  2. L13:2594 suppresses
  3. L23:2666 weakly excites
  4. L7:1078 weakly excites
  5. L18:628 weakly suppresses
  6. L7:373 weakly suppresses
  7. L22:1562 weakly excites
  8. L25:1468 weakly excites
  9. L11:797 weakly suppresses
  10. L9:2299 weakly excites
  11. L12:2307 weakly excites
  12. L9:2569 weakly excites
  13. L24:2778 weakly excites
  14. L25:2303 weakly excites
  15. L6:3061 weakly excites
  16. L16:1915 weakly excites
  17. L16:2970 weakly excites
  18. L8:2647 weakly excites
  19. L6:2862 weakly suppresses
  20. L9:2545 weakly excites
  21. L6:54 weakly excites
  22. L9:1455 weakly excites
  23. L23:2451 weakly excites
  24. L9:847 weakly 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 -1.85 and -1.741 values between -1.74 and -1.631 values between -1.31 and -1.211 values between -1.21 and -1.103 values between -0.89 and -0.784 values between -0.78 and -0.684 values between -0.68 and -0.575 values between -0.57 and -0.4611 values between -0.46 and -0.3655 values between -0.36 and -0.25273 values between -0.25 and -0.14953 values between -0.14 and -0.041676 values between -0.04 and 0.07644 values between 0.07 and 0.17209 values between 0.17 and 0.2839 values between 0.28 and 0.3912 values between 0.39 and 0.496 values between 0.49 and 0.602 values between 0.60 and 0.712 values between 0.71 and 0.813 values between 0.81 and 0.921 values between 0.92 and 1.031 values between 1.03 and 1.131 values between 2.30 and 2.41offnegativepositiveoften

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. lée
  2. @student
  3. MITTED
  4. peri
  5. utschen
  6. atedRoute
  7. TERNAL
  8. виз
  9. meno
  10. utation

Suppresses

  1. olar
  2. ouser
  3. contains
  4. constraints
  5. ��
  6. トー
  7. جيب
  8. ideals
  9. -quarter
  10. zaw