Neuron Explorer

Qwen3-0.6B

first digit of a split number

A neuron in layer 7 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 a single digit, usually a 1 or a 2, that starts a larger number, such as the first digit in '1902' or '2.04' when the number is cut into pieces. Turning it up made the model write long lists of numbers, which fits the same theme.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    that deepens their knowledge while also developing 21st-century skills.⏎Read the rest of this entry »⏎Filed under: Gaming, Services, Software, 3 million dollar grant, Associate Professor, Augmented reality, Bill & Melinda

  2. #2 peak

    of unzipping the .tgz files using the terminal. Let’s get into these.⏎Note: For this particular tutorial, we will use the Ubuntu 22.04. However, the commands used here are same for all the

  3. #3 peak

    It was discovered in 1787 by William Herschel. The star SAO 20575 or BD+602522 is thought to

  4. #4 peak

    atta Facts⏎Mohammad Hatta (1902-1980), one of the foremost intellectuals in the

  5. #5 peak

    elage of the Nieuport 12 with an 80hp Le Rhône engine. The Nieu

  6. #6 peak

    JSP) environment, marrying XML data to Java-based services and applications is easier than ever before. The September 17 release of the JavaServer Pages 1.2 specification (⏎- The new specification includes better mechanisms for accessing and describing XML

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: it is a good idea. The situation is in the current state of being in the storm of... a good idea. Now it is in the storm of... a good idea. I am not sure. The situation is in the storm of

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

    S ONE 8 (2): e57075. . . http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.005

  2. #2 peak

    345 Can we simplify? No, that's not possible.⏎The answer is 315 + 35 = 345⏎Example 2: 214

  3. #3 peak

    to Earth is the Sun. All other stars are millions of miles away. Scientists estimate that the entire Universe (composed of billions of galaxies) has more than 10 billion trillion stars.⏎From our earthly perspective, stars appear very

  4. #4 peak

    300 in grant money to help him develop a more effective way of teaching math to fourth graders. He used it to buy a Wii.⏎The $300 came to the Bacon Elementary School teacher as part of a special science and

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. L4:53 excites blank slot before a number
  2. L4:448 excites
  3. L4:205 weakly excites the digit 2 inside citations
  4. L6:29 weakly suppresses
  5. L3:1099 weakly suppresses
  6. L2:83 weakly suppresses
  7. L2:2896 weakly excites
  8. L3:967 weakly excites
  9. L5:1136 weakly excites
  10. L5:5 weakly excites digit after a label like paper or type
  11. L3:2166 weakly suppresses
  12. L0:99 weakly excites
  13. L0:529 weakly excites
  14. L1:551 weakly suppresses
  15. L5:874 weakly suppresses
  16. L3:315 weakly suppresses 'of' after a noun phrase
  17. L0:1145 weakly excites
  18. L4:886 weakly excites
  19. L5:2253 weakly suppresses
  20. L2:2070 weakly suppresses
  21. L1:2671 weakly excites inside multi-part proper names
  22. L1:1888 weakly suppresses
  23. L4:995 weakly suppresses
  24. L4:326 weakly suppresses

Feeds later layers

  1. L17:157 excites
  2. L10:395 excites
  3. L11:15 excites decade and century numbers like 1950s
  4. L10:600 excites
  5. L17:285 excites
  6. L13:1257 suppresses
  7. L18:122 excites
  8. L19:298 excites
  9. L17:128 excites
  10. L18:1412 excites
  11. L15:828 excites
  12. L9:13 excites
  13. L10:294 excites
  14. L17:1580 excites
  15. L10:1081 excites
  16. L10:50 excites
  17. L24:1726 excites
  18. L20:1100 excites
  19. L19:723 excites
  20. L16:35 suppresses 'not' in 'but not limited to'
  21. L16:372 excites
  22. L14:2281 excites
  23. L20:729 excites
  24. L19:785 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 -2.26 and -2.121 values between -1.99 and -1.861 values between -1.72 and -1.593 values between -1.59 and -1.451 values between -1.32 and -1.192 values between -1.19 and -1.051 values between -1.05 and -0.922 values between -0.92 and -0.791 values between -0.79 and -0.6512 values between -0.65 and -0.5231 values between -0.52 and -0.39162 values between -0.39 and -0.25672 values between -0.25 and -0.121799 values between -0.12 and 0.02927 values between 0.02 and 0.15177 values between 0.15 and 0.2835 values between 0.28 and 0.4213 values between 0.42 and 0.5514 values between 0.55 and 0.689 values between 0.68 and 0.827 values between 0.82 and 0.959 values between 0.95 and 1.084 values between 1.08 and 1.222 values between 1.22 and 1.357 values between 1.35 and 1.492 values between 1.49 and 1.623 values between 1.62 and 1.751 values between 1.75 and 1.891 values between 1.89 and 2.022 values between 2.02 and 2.151 values between 2.15 and 2.293 values between 2.42 and 2.551 values between 2.69 and 2.821 values between 2.95 and 3.09offnegativepositiveoften

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. 龄
  2. 齡
  3. 岁以下
  4. eenth
  5. 岁以上
  6. orget
  7. 季后
  8. olon
  9. Nin
  10. 疗程

Suppresses

  1. symbols
  2. 符号
  3. tas
  4. aped
  5. ap
  6. 物质
  7. symbol
  8. apl
  9. project
  10. iant