Neuron Explorer

Qwen3-0.6B

'such' introducing examples

A neuron in layer 21 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 lights up on the word 'such' when it introduces examples, as in 'things such as books'. It is weakly negative on 'first' in 'for the first time' and on 'so' in 'and so on'. Turning it up produced unrelated symbol patterns, so the steering does not show this 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

    medication in addition to other causes.⏎Here are some of those reasons in detail:⏎1. Fear or Anxiety⏎Anxiety in dogs can come in many forms, such as separation anxiety, generalized anxiety, environmental

  2. #2 peak

    posed of transparent, overlapping scales, it resembles the shingles on a roof. The cuticle's primary purpose is to safeguard the inner structures of the hair, such as the cortex and medulla, from damage caused by external factors like heat,

  3. #3 peak

    it is EPA-registered to kill the influenza A virus and norovirus.⏎- Disinfect high-touch surfaces often. Objects and surfaces that are frequently touched such as desks, countertops, doorknobs, keyboards, faucets, and

  4. #4 peak

    <|endoftext|>Whether it’s a child or a teenager, a lot of us have a backpack to use when going to school. These backpacks are used to store things such as books, notebooks, and

  5. #5 peak

    components of a bilan comptable is essential. This comprehensive guide aims to provide a clear understanding of what a bilan comptable entails, including its key components such as assets, liabilities, and equity. By delving into the importance of the

  6. #6 peak

    the chest. Severe carbon monoxide poisoning can cause neurological damage, coma and death.⏎Sources of carbon monoxide can include anything that uses combustion to operate, such as gas generators, power tools, compressors, pumps, welding equipment, space

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 first year a person with a very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very very

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

    zzurri to their second consecutive world title, as well as an Olympic gold medal in 1936.⏎Asia was represented in the tournament for the first time as Dutch East Indies, now Indonesia, qualified while other Asian teams withdrew.⏎

  2. #2 peak

    people with disabilities. You may also ask if the practitioner makes house calls.⏎Your file contains<|endoftext|>The United States just won the gold medal in curling for the first time in history. You might be wondering what this strange sport is all about.

  3. #3 peak

    servos, etc., but it is limited and doesn't provide what an OS provides, e.g., multiple processes or threads of execution, signals, IPC and so on.⏎IPC - In computer science, 'Inter Process Communication' (IPC)

  4. #4 peak

    aceans related to shrimp and crabs) and sea spiders.⏎The recent collapse of several Antarctic Peninsula ice shelves has exposed seafloor communities to light for the first time. An expedition to the disintegrating Larsen A and B ice shelves

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. L20:1114 strongly excites
  2. L20:1043 suppresses
  3. L18:928 suppresses
  4. L18:212 suppresses
  5. L19:1463 excites
  6. L20:1860 suppresses
  7. L19:1521 excites
  8. L20:858 suppresses
  9. L20:1229 suppresses
  10. L19:2626 excites
  11. L20:95 suppresses
  12. L10:346 suppresses "each" in "each other" phrases
  13. L20:80 suppresses
  14. L18:669 excites
  15. L9:929 excites 'each' in 'each other'
  16. L17:1002 excites
  17. L20:215 suppresses
  18. L20:3066 excites
  19. L7:663 suppresses
  20. L20:67 suppresses
  21. L20:457 excites
  22. L19:1871 suppresses
  23. L19:2446 suppresses
  24. L19:2565 suppresses

Feeds later layers

  1. L24:85 suppresses
  2. L22:225 suppresses
  3. L22:238 excites
  4. L24:48 suppresses
  5. L26:26 suppresses
  6. L25:420 excites
  7. L22:201 weakly excites
  8. L22:29 weakly suppresses
  9. L25:3 weakly suppresses
  10. L22:209 weakly excites
  11. L23:263 weakly excites
  12. L22:175 weakly excites
  13. L25:35 weakly suppresses
  14. L24:363 weakly excites
  15. L22:498 weakly excites
  16. L25:5 weakly suppresses
  17. L27:857 weakly excites
  18. L22:1317 weakly excites unclear: sentence and clause breaks in essays
  19. L25:1255 weakly excites
  20. L22:4 weakly excites
  21. L23:390 weakly excites
  22. L27:723 weakly excites
  23. L22:0 weakly excites
  24. L22:345 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.

2 values between -5.49 and -5.082 values between -4.67 and -4.272 values between -4.27 and -3.868 values between -3.45 and -3.059 values between -3.05 and -2.648 values between -2.64 and -2.2324 values between -2.23 and -1.8335 values between -1.83 and -1.4257 values between -1.42 and -1.01115 values between -1.01 and -0.61602 values between -0.61 and -0.202220 values between -0.20 and 0.21493 values between 0.21 and 0.62125 values between 0.62 and 1.0273 values between 1.02 and 1.4350 values between 1.43 and 1.8422 values between 1.84 and 2.2416 values between 2.24 and 2.6512 values between 2.65 and 3.0614 values between 3.06 and 3.469 values between 3.46 and 3.873 values between 3.87 and 4.282 values between 4.69 and 5.092 values between 5.91 and 6.311 values between 7.53 and 7.942 values between 10.38 and 10.79offnegativepositiveoften

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. ice
  2. etyl
  3. ear
  4. ete
  5. хват
  6. ONG
  7. ola
  8. од
  9. axter
  10. esty

Suppresses

  1. 才是真正
  2. orary
  3. 了大量的
  4. 也是一个
  5. 性强
  6. (BuildContext
  7. 部副
  8. ="/">⏎
  9. 工作经验
  10. 并不意味