Neuron Explorer

Qwen3-0.6B

'host' or 'number' before 'of'

A neuron in layer 5 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 word that comes after 'a' in phrases like 'a whole host of' and 'a number of', right before the word 'of'. It goes quiet or negative where a percentage or figure should follow 'a', as in 'up to a 90 percent cut'. Turning it up made the model fall into repeated 'of the' phrasing.

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

    .⏎Water is the essence of life, but not just human life. Almost every living thing on the planet must have water to survive, and that includes a whole host of bacteria and microorganisms that can wreak havoc on human health.

  2. #2 peak

    hygiene. Read on to learn more.⏎Negative Effects of Poor Sleep⏎Aside from feeling cranky, tired and generally out of sorts, sleep deprivation can cause a host of physical issues. Loss of sleep can affect everything from short-term memory to one

  3. #3 peak

    many petals then had fallen?<|endoftext|>At EdPlace we are passionate about planning and preparation to avoid undue stress to students, whatever their age.⏎There are a whole host of reasons why a 6-week learning plan may be useful to your child and

  4. #4 peak

    that is not reflected by any Latin character. In such cases, combination of Latin characters are often used to represent such sound.⏎- Macro Language⏎There are a number of different Chinese “Dialects”, some which differ so much from each other

  5. #5 peak

    and third, a significant nummber of Christians who became later targets. Such a combination of factors did not exist anywhere

  6. #6 peak

    Unit 13 1. 1 Describe the different types of documents that may be produced from the notes and the formats to be followed. There are a number of different types of documents, below is a table which shows the types of documents

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: I have a 3250 word problem in this question. You can check this. Answer: The answer is "How to find 3250 words." Please find this. Answer: The user uploaded an image with this

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

    ner’s grandparents, aunt and uncles. If he hadn’t been able to escape to the United States, Goldner says he would have had a less than 7 percent chance of survival in Germany.⏎Jessica Sternbach, president of the Jewish

  2. #2 peak

    2.7 m3 per MWh, but wind power uses only a fraction of these amounts.⏎In the US, the Department of Energy estimates that with a 20% share of wind power in

  3. #3 peak

    change emissions by up to 30 percent, according to the Portland Bureau of Planning and Sustainability. RNG’s climate benefits are even greater, offering up to a 90 percent cut in climate pollution.⏎The term “renewable” refers

  4. #4 peak

    the daily value for saturated fat. Eating too much total fat and saturated fat elevates your blood cholesterol levels and increases your risk for heart disease. Also, a 100-gram serving of tamales delivers approximately 3.1

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. L3:2559 suppresses
  2. L4:392 suppresses
  3. L4:49 excites
  4. L4:764 weakly suppresses
  5. L2:1003 weakly suppresses
  6. L1:740 weakly suppresses
  7. L3:858 weakly suppresses
  8. L4:1537 weakly suppresses
  9. L2:533 weakly excites
  10. L4:21 weakly excites
  11. L2:493 weakly suppresses
  12. L2:984 weakly suppresses
  13. L2:851 weakly suppresses
  14. L2:501 weakly excites
  15. L4:367 weakly suppresses
  16. L3:286 weakly excites
  17. L0:2008 weakly suppresses
  18. L2:579 weakly suppresses
  19. L3:272 weakly excites
  20. L0:161 weakly suppresses
  21. L4:229 weakly excites
  22. L1:427 weakly excites
  23. L1:951 weakly excites
  24. L4:19 weakly suppresses

Feeds later layers

  1. L7:224 strongly excites
  2. L8:1021 excites
  3. L8:202 excites
  4. L8:1029 excites
  5. L6:1588 excites
  6. L12:1561 excites
  7. L8:1368 excites
  8. L20:2519 excites "of" after quantity nouns (millions of, amount of)
  9. L10:1465 excites
  10. L10:707 excites
  11. L15:2739 weakly excites
  12. L22:142 weakly suppresses 'of' after a quantity or noun phrase
  13. L17:1246 weakly excites
  14. L10:1153 weakly excites
  15. L16:1009 weakly excites
  16. L8:678 weakly excites
  17. L17:1516 weakly excites
  18. L7:1524 weakly excites
  19. L15:1896 weakly suppresses
  20. L14:1108 weakly suppresses
  21. L25:1679 weakly excites
  22. L13:1273 weakly suppresses
  23. L6:586 weakly excites
  24. L6:325 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 -3.94 and -3.691 values between -3.69 and -3.431 values between -2.66 and -2.401 values between -1.88 and -1.631 values between -1.63 and -1.372 values between -1.37 and -1.115 values between -1.11 and -0.864 values between -0.86 and -0.6025 values between -0.60 and -0.34675 values between -0.34 and -0.082946 values between -0.08 and 0.18221 values between 0.18 and 0.438 values between 0.43 and 0.695 values between 0.69 and 0.952 values between 0.95 and 1.203 values between 1.20 and 1.462 values between 1.46 and 1.721 values between 1.72 and 1.981 values between 2.49 and 2.751 values between 3.26 and 3.521 values between 3.52 and 3.781 values between 6.09 and 6.35offnegativepositiveoften

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. rac
  2. 人群中
  3. ued
  4. ensed
  5. quarters
  6. uetooth
  7. ながら
  8. 人群
  9. quarters
  10. bien

Suppresses

  1. inem
  2. 在玩家
  3. administrator
  4. theirs
  5. .Actions
  6. .setState
  7. .DataBindings
  8. .googleapis
  9. designers
  10. 您的