Neuron Explorer

Qwen3-0.6B

"a" meaning per (a day, a year)

A neuron in layer 9 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 "a" when it means "per", as in "100 people a day" or "once a year". It is quiet, and somewhat negative, on "a" in other places like "sign up for a prep course". Turning it up did not give a clear change in the output.

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

    . Symptoms usually include fever, pain, swelling, redness, pus, runny nose, sore throat, or swollen lymph nodes.⏎Possible responses:→ Once a year or

  2. #2 peak

    according to the World Health Organization (WHO). In the United States 110 people a day die on our roads and highways, that’s one death every 13

  3. #3 peak

    izzly charge. The bears were obviously aware of my presence, but upwind, and in the dark, they were unable to locate my exact position. After over a minute of intense sniffing and scanning, the large sow eventually decided to saunter

  4. #4 peak

    in 1856 called the London general Omnibus Company and it soon became the largest bus company in the world carrying around 100 million passengers a year in 1890.⏎Horse trams, can also be

  5. #5 peak

    so a tool isn’t always calling the shots. (We get enough of that in politics, yes?)⏎It’s super easy to get started too.⏎Here’s a quick how-to:⏎- Get in a comfy position. You don’t need to

  6. #6 peak

    putting the major burden of the wars on that class.⏎We’ve answered 319,210 questions. We can answer yours, too.Ask a question<|endoftext|>Given how important financial skills are to navigating life, it’s surprising that

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: a new project. The project that I have a small 100000000000000000000000000000000000

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

    it to see if it makes sense. Students who want to take advantage of these and other strategies for the GRE are encouraged to contact our offices to sign up for a prep course today.⏎Read Magazines and Newspapers⏎Reading magazines and newspapers is

  2. #2 peak

    a rectangular concrete wall or roof panel is among the most common form of precast concrete products provided by concrete<|endoftext|>Welcome to Word of the Week! Stay tuned for a new word each Friday to amp up your nature vocabulary!⏎Denudate [d

  3. #3 peak

    of chaotic systems (H. Qammari, A. Venkatesan). A vacation on Mars - an artist's journey in

  4. #4 peak

    players were moving their piece according to the result of the throw of a special 7-sided die. They were winning or losing stakes depending on the formed

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. L4:137 strongly excites
  2. L4:49 strongly excites
  3. L3:1810 excites
  4. L1:1062 suppresses
  5. L0:173 excites
  6. L4:731 excites
  7. L7:224 suppresses
  8. L3:914 excites "any" in negated sentences
  9. L8:1673 suppresses
  10. L4:2589 suppresses
  11. L2:1829 suppresses
  12. L6:426 excites
  13. L8:10 excites
  14. L4:961 excites
  15. L8:466 suppresses 'a' or 'an' before a noun phrase
  16. L4:73 excites
  17. L8:7 excites
  18. L3:2559 weakly suppresses
  19. L5:457 weakly suppresses 'a' after 'tell', 'bring' or 'much of'
  20. L5:225 weakly suppresses
  21. L8:13 weakly suppresses
  22. L8:3 weakly excites
  23. L8:4 weakly excites
  24. L5:1416 weakly suppresses

Feeds later layers

  1. L11:331 strongly suppresses
  2. L16:2248 strongly suppresses
  3. L10:365 strongly suppresses
  4. L17:188 strongly suppresses 'a' before a noun phrase
  5. L16:1309 strongly suppresses
  6. L13:824 suppresses
  7. L18:1477 suppresses
  8. L20:901 suppresses
  9. L17:1179 suppresses
  10. L21:592 suppresses
  11. L11:875 excites
  12. L16:880 suppresses
  13. L17:1542 suppresses
  14. L18:314 excites
  15. L15:869 suppresses
  16. L10:1162 suppresses
  17. L19:2088 suppresses
  18. L18:1084 suppresses
  19. L18:2389 suppresses
  20. L16:2062 excites "other" and contrast words
  21. L22:1881 suppresses
  22. L18:2608 suppresses
  23. L13:1234 excites
  24. L16:1212 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 -10.36 and -10.051 values between -9.13 and -8.821 values between -8.82 and -8.511 values between -8.20 and -7.891 values between -6.97 and -6.661 values between -6.66 and -6.352 values between -6.35 and -6.041 values between -6.04 and -5.742 values between -5.74 and -5.435 values between -5.43 and -5.128 values between -5.12 and -4.813 values between -4.81 and -4.506 values between -4.50 and -4.195 values between -4.19 and -3.892 values between -3.89 and -3.589 values between -3.58 and -3.2712 values between -3.27 and -2.968 values between -2.96 and -2.655 values between -2.65 and -2.3511 values between -2.35 and -2.046 values between -2.04 and -1.7314 values between -1.73 and -1.4210 values between -1.42 and -1.1112 values between -1.11 and -0.8018 values between -0.80 and -0.5043 values between -0.50 and -0.19708 values between -0.19 and 0.122702 values between 0.12 and 0.43302 values between 0.43 and 0.746 values between 0.74 and 1.052 values between 1.66 and 1.97offnegativepositiveoften

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. 得益
  2. ments
  3. stages
  4. ponents
  5. generations
  6. eye
  7. 一审
  8. m
  9. 效果
  10. 作用

Suppresses

  1. lec
  2. ustralian
  3. )[
  4. 全新的
  5. лик
  6. particular
  7. 特殊的
  8. ustin
  9. inte
  10. 不一样的