Neuron Explorer

Qwen3-0.6B

unclear: hyphens and size comparisons

A neuron in layer 6 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron is hard to pin down. It fires most on hyphens inside compound words and on phrases that compare sizes, such as 'twice the size' or 'times the height'. The steered text did not show a clear pattern.

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

    the speeds of two satellites placed in stable circular orbits of radius R and 2R respectively?⏎C) 1 / 2⏎D) 1 / �

  2. #2 peak

    refer to their objects⏎> through a Map. Then changing a leaf object requires only that the value⏎> associated with key represented by the leaf be (re-)inserted into the⏎> Map. The Tree itself need not change at all

  3. #3 peak

    for example, in the controversy concerning the effect of electromagnetic fields on health, the field strengths have almost always been reported in milligauss.⏎The mechanical units are the same in all cgs systems. Length,

  4. #4 peak

    designed to support orbits for the satellites, to perform orbit control tasks for small-, mini- and micro-satellites.<|endoftext|>Until 1864, Georgia had

  5. #5 peak

    technologies, databases on the fundamental properties of “nano-structured” materials, and other essential supporting tools and capabilities. U.S. industry

  6. #6 peak

    North Pacific Gyre.⏎Twice the size of Texas – though some scientists have argued that its mass has extended to twice

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: "What is the purpose of the first part of the sentence, 'To the best of my knowledge' I am able to achieve' (A. P. S. I.)" I will do it, the purpose of the first part

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

    the expressions of V(t), I(t), the trigonometric formula sin(X)sin(Y)=1/2(cos(X-Y)-cos(X+Y)), and the

  2. #2 peak

    power is then written as⏎Here the average of sin2 ωt over a cycle is 1/2 and that of sin ω t cos ωt is zero. Substituting these values, we obtain average power

  3. #3 peak

    9 Copyright Math Open Reference. All rights reserved<|endoftext|>Use chain rule⏎F'(x) = dF(x)/dx = (dF(x)/dx2)(dx2/dx) = 2x (dF(x)/dx2

  4. #4 peak

    the input-output pair. It is called supervised learning because the process of an algorithm learning from the

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. L5:33 excites
  2. L5:227 weakly suppresses
  3. L5:27 weakly excites
  4. L5:219 weakly suppresses
  5. L3:286 weakly excites
  6. L5:75 weakly excites
  7. L4:493 weakly suppresses
  8. L4:592 weakly excites
  9. L2:2796 weakly suppresses
  10. L3:2140 weakly suppresses
  11. L2:359 weakly suppresses
  12. L4:522 weakly suppresses
  13. L4:97 weakly suppresses
  14. L1:146 weakly excites
  15. L5:1833 weakly suppresses
  16. L4:896 weakly excites
  17. L3:338 weakly suppresses
  18. L4:458 weakly excites
  19. L5:2234 weakly suppresses
  20. L2:1684 weakly excites
  21. L4:25 weakly excites
  22. L4:544 weakly excites
  23. L4:1908 weakly excites
  24. L0:758 weakly suppresses

Feeds later layers

  1. L16:1275 suppresses
  2. L19:789 weakly suppresses
  3. L9:335 weakly excites 'of' after nouns like anniversary or onset
  4. L9:1854 weakly suppresses
  5. L17:101 weakly suppresses
  6. L18:2385 weakly suppresses
  7. L10:1018 weakly excites
  8. L17:2764 weakly excites
  9. L16:1216 weakly suppresses
  10. L7:2611 weakly excites
  11. L23:1542 weakly suppresses
  12. L14:636 weakly excites
  13. L20:3031 weakly suppresses
  14. L9:2462 weakly excites
  15. L15:1162 weakly suppresses
  16. L12:1896 weakly excites
  17. L11:1431 weakly suppresses commas and 'and' in noun lists
  18. L10:1568 weakly excites
  19. L7:2164 weakly excites period inside country abbreviations
  20. L12:277 weakly suppresses
  21. L14:319 weakly suppresses
  22. L18:1492 weakly suppresses
  23. L12:2959 weakly excites
  24. L7:1031 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.

3 values between -0.65 and -0.601 values between -0.55 and -0.501 values between -0.50 and -0.454 values between -0.45 and -0.4012 values between -0.40 and -0.3525 values between -0.35 and -0.3062 values between -0.30 and -0.25125 values between -0.25 and -0.21187 values between -0.21 and -0.16311 values between -0.16 and -0.11462 values between -0.11 and -0.06613 values between -0.06 and -0.01767 values between -0.01 and 0.04485 values between 0.04 and 0.09339 values between 0.09 and 0.14204 values between 0.14 and 0.19131 values between 0.19 and 0.2494 values between 0.24 and 0.2939 values between 0.29 and 0.3317 values between 0.33 and 0.3813 values between 0.38 and 0.434 values between 0.43 and 0.484 values between 0.48 and 0.531 values between 0.58 and 0.631 values between 0.63 and 0.681 values between 0.68 and 0.731 values between 0.92 and 0.971 values between 1.26 and 1.31offnegativepositiveoften

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. loid
  2. cep
  3. legit
  4. aran
  5. objectId
  6. avel
  7. arkan
  8. eland
  9. nexus
  10. +f

Suppresses

  1. itivity
  2. devote
  3. ;⏎
  4. 隳
  5. ;")⏎
  6. .";⏎
  7. ("-");⏎
  8. 同事们
  9. --*/⏎
  10. ⏎ ⏎