Neuron Explorer

Qwen3-0.6B

gap before a year in dates

A neuron in layer 12 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 space right after a month and day, like 'born on August 1,', where the next thing is a year. When it was turned up, the model tended to write years such as 2005, 1929 and 2020.

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

    .⏎Yves Saint Laurent was a European fashion designer born on August 1, 1936 in Oran, Algeria. As a teen, he left

  2. #2 peak

    a charge of bad taste,<|endoftext|>8 edition of Principles of Biblical Interpretation found in the catalog.⏎April 1, 2003⏎by Baker Academic⏎Written in English⏎|The Physical

  3. #3 peak

    |⏎|Born||December 7, 1928⏎|Era||20th / 21

  4. #4 peak

    “thermostatic mechanism” that would counter-balance the dominant presence of electronic media in the lives of children.⏎How to Cite⏎Copyright (c) 2017 Robert Albrecht⏎Copyrights and publishing rights of all

  5. #5 peak

    province of Segovia, in 1559; died at Madrid, 27 March, 1625. He was a great-grandson of the Tordesillas

  6. #6 peak

    of the most beloved leaders India has ever produced.⏎Life history of Subhash Chandra Bose⏎Subhash Chandra Bose was born on 23 January, 1897 in Cuttack (Orissa) to Janakinath

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: and I have, but I have, I have, I have, I have, I have, I have, I have, I have, I have, I have, I have, I have, I have, I have, I

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

    mynursingpaper.com/wp-content/uploads/2020/06/cropped-1559720319

  2. #2 peak

    rights to unionized or lack thereof).Please provide resources as required by APA standards.⏎https://mynursingpaper.com/wp-content/uploads/2020/06/cropped-1559720319

  3. #3 peak

    data. However, a means of observing the kinds of relationships between variables was needed. Using the theory of linear regression which originated from studies performed by Sir Francis Galton (1822-1911), the scatter diagram was developed

  4. #4 peak

    servant” are perhaps better understood as “slave,” in our times not an acceptable self-identity. Another difficulty: the actual lifestyles of servants/slaves in ancient cultures. Slaves were engaged by rich people to carry out their commands. Serv

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. L11:2530 strongly excites
  2. L6:2811 suppresses
  3. L9:1460 excites
  4. L11:1975 suppresses
  5. L11:2229 excites
  6. L9:811 suppresses
  7. L9:1880 suppresses
  8. L8:457 excites year digits in copyright and dates
  9. L4:1291 excites second digit of years 20xx
  10. L4:1395 weakly suppresses
  11. L11:464 weakly suppresses
  12. L10:456 weakly suppresses
  13. L9:540 weakly suppresses
  14. L9:955 weakly excites
  15. L8:2335 weakly excites
  16. L9:534 weakly excites
  17. L9:1311 weakly suppresses
  18. L4:2321 weakly excites
  19. L9:382 weakly suppresses
  20. L8:2347 weakly suppresses
  21. L9:2452 weakly suppresses
  22. L11:1865 weakly excites open parenthesis in citations
  23. L1:2444 weakly suppresses
  24. L8:534 weakly suppresses

Feeds later layers

  1. L15:2236 excites
  2. L17:1476 excites
  3. L13:2285 excites
  4. L18:2103 excites
  5. L13:1342 excites digits inside years like 2017
  6. L15:1010 excites year digits inside 20xx dates
  7. L16:394 excites
  8. L18:1984 excites
  9. L14:1253 excites
  10. L17:1264 excites
  11. L18:2542 excites
  12. L14:1803 excites
  13. L19:1329 excites
  14. L14:1700 excites
  15. L18:1865 weakly excites
  16. L14:2374 weakly suppresses
  17. L19:2030 weakly excites
  18. L22:1279 weakly excites
  19. L14:809 weakly excites
  20. L25:2960 weakly excites
  21. L13:376 weakly excites
  22. L13:1322 weakly excites
  23. L15:622 weakly suppresses
  24. L14:2439 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.

3 values between -1.32 and -0.9623 values between -0.96 and -0.61247 values between -0.61 and -0.252883 values between -0.25 and 0.11673 values between 0.11 and 0.4716 values between 0.47 and 0.827 values between 0.82 and 1.1810 values between 1.18 and 1.544 values between 1.54 and 1.893 values between 1.89 and 2.255 values between 2.25 and 2.611 values between 2.61 and 2.962 values between 2.96 and 3.322 values between 3.32 and 3.681 values between 3.68 and 4.041 values between 4.04 and 4.392 values between 4.75 and 5.111 values between 5.11 and 5.462 values between 5.46 and 5.822 values between 5.82 and 6.181 values between 6.18 and 6.534 values between 6.53 and 6.892 values between 6.89 and 7.251 values between 7.61 and 7.962 values between 7.96 and 8.322 values between 8.68 and 9.032 values between 9.03 and 9.391 values between 9.39 and 9.752 values between 10.11 and 10.461 values between 10.46 and 10.821 values between 11.53 and 11.891 values between 12.60 and 12.96offnegativepositiveoften

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. 公元
  2. WXYZ
  3. 晚年
  4. 垓
  5. 近几年
  6. 公元前
  7. recent
  8. getObject
  9. 新的一年
  10. abis

Suppresses

  1. 指数
  2. .ct
  3. AGR
  4. 委员
  5. 篪
  6. rippling
  7. Indicator
  8. ăng
  9. cong
  10. hart