Neuron Explorer

Qwen3-0.6B

unclear: weak noun and pronoun activity

A neuron in layer 21 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

Its signals are weak and do not form a clear pattern. It loosely fires on nouns like 'years', 'seed' and 'music' and goes quiet on words like 'time' and 'ultimately'. Turning it up or down gave repetitive text, so it does not show a clear job.

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 results are still not perfect, and there is still much work to be done, but the technology is very promising, and we hope that in a few years we will have systems that can completely break down language barriers ... We may

  2. #2 peak

    ery as a flower from a seed. No sooner did the soft, sweet note of the bow-string charm the ear of genius than music was born, and from music came poetry and painting and sculpture.⏎What the fragments of Sapphic song and

  3. #3 peak

    4-5), “I will make your seed to multiply as the stars of heaven, and I will give to your seed all these countries. And in your seed shall all the nations of the earth be blessed because Abraham obeyed my voice,

  4. #4 peak

    īvā- “the back part of the neck, nape, neck,” see Monier-Williams, p. 374; on this subject and on the etymology of further related Indo-European references which, among other things

  5. #5 peak

    for a potential outbreak of measles. Officials in Herefordshire fear the epidemic in Wales means an outbreak is 'on the doorstep'.⏎They say that so far this year they've already had the same number of cases as they had for the whole of

  6. #6 peak

    from Dakar. Issa the Fulani, from Senegal’s brushy hinterlands. Moustapha, most recently from Spain, where in five years he could

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: , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,

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 terminal circuit is complete.⏎Batteries work via a combination of oxidation and reduction. The anode oxidizes, meaning it loses electrons. At the same time, the cathode absorbs electrons in a process called reduction.

  2. #2 peak

    entangled. The research also integrates the materials with chip-scale silicon photonic devices that confine the light to extremely small volume to enhance the light emission efficiency. Ultimately, this research provides significant insight into how to control the properties of atomically thin

  3. #3 peak

    And Resolutions can be anything from eating healthier to spending more time with family. Resolutions may not have any specific timeline or plan of action associated with them⏎Ultimately, both goals and resolutions can help individuals reach their desired outcomes; however, it

  4. #4 peak

    PCI Security Standards Council. Outdated browsers do not support the requirements of modern web applications and pose significant security risks. After May 1, 2017, this browser will no longer be supported and you may experience issues or not be

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. L20:660 strongly excites
  2. L17:1845 excites
  3. L20:122 excites
  4. L19:1972 suppresses
  5. L17:2397 excites
  6. L20:2430 excites
  7. L18:160 suppresses
  8. L18:1791 suppresses
  9. L17:1426 excites
  10. L18:843 excites
  11. L13:1232 excites
  12. L11:237 suppresses
  13. L19:1084 suppresses "and" joining clauses after a condition
  14. L20:822 excites
  15. L20:2098 excites
  16. L16:1838 excites
  17. L20:132 excites
  18. L20:1312 suppresses
  19. L20:2891 suppresses
  20. L18:2996 suppresses
  21. L18:862 suppresses hedges like 'it turns out' and 'if any'
  22. L20:577 excites
  23. L15:1792 excites
  24. L20:277 suppresses

Feeds later layers

  1. L22:1124 strongly excites
  2. L23:298 excites
  3. L27:717 excites
  4. L22:360 excites
  5. L26:236 suppresses
  6. L22:689 excites comma closing an intro clause
  7. L26:715 excites
  8. L24:216 suppresses
  9. L27:86 excites
  10. L22:353 suppresses
  11. L22:368 suppresses
  12. L25:1221 suppresses
  13. L26:22 excites the 'res' fragment inside words
  14. L22:892 excites
  15. L25:247 excites
  16. L22:940 suppresses
  17. L27:108 suppresses
  18. L26:2123 suppresses
  19. L24:572 suppresses
  20. L25:234 excites
  21. L27:358 suppresses
  22. L24:1366 excites
  23. L26:693 weakly excites
  24. L27:286 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 -7.72 and -7.311 values between -7.31 and -6.901 values between -4.86 and -4.451 values between -4.04 and -3.631 values between -3.22 and -2.812 values between -2.81 and -2.404 values between -2.40 and -2.003 values between -2.00 and -1.591 values between -1.59 and -1.1816 values between -1.18 and -0.77359 values between -0.77 and -0.362245 values between -0.36 and 0.051145 values between 0.05 and 0.4670 values between 0.46 and 0.8616 values between 0.86 and 1.2715 values between 1.27 and 1.686 values between 1.68 and 2.098 values between 2.09 and 2.504 values between 2.50 and 2.911 values between 2.91 and 3.321 values between 3.32 and 3.722 values between 3.72 and 4.132 values between 4.13 and 4.541 values between 4.54 and 4.951 values between 5.77 and 6.171 values between 8.22 and 8.63offnegativepositiveoften

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. 我把
  2. they
  3. we
  4. 我
  5. you
  6. 我还是
  7. świadczenie
  8. 我发现
  9. 你
  10. everyone

Suppresses

  1. ,
  2. ,body
  3. ,
  4. ,en
  5. ,N
  6. ,U
  7. ،
  8. ,function
  9. ,Z
  10. ,M