Neuron Explorer

Qwen3-0.6B

timestamps after dates

A neuron in layer 3 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 space right after a date, just before a clock time such as '08:57' or '23:34:59'. Turning it up pushed the model toward ordinary prose and question-like text, not toward dates.

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

    , osteoporosis, meniscus, ligament injuries, muscle problems.⏎Günceleme: 07/01/2023 14:52<|endoftext|>Sustainability and Renewable Merino⏎Wool runs

  2. #2 peak

    7.png admin2021-10-20 23:34:592021-10-20 23:34:59assignment 2⏎mynursing

  3. #3 peak

    Jan 09, 2017 08:57 AM EST⏎MIT Professor Is Building Bio-Inspired

  4. #4 peak

    would be removed due merely to the<|endoftext|>Editorial: Team effort needed to combat cyberbullying⏎Updated On: Sep 21 2012 11:49:55 AM EDT⏎This is a weekly editorial

  5. #5 peak

    (nOPC=0 and MAS[1:0]=10).⏎Article last edited on: 2008-09-09 15:47:35⏎Did you find this article helpful?

  6. #6 peak

    Sep 28, 2009 8:19 PM by Michael Marsh⏎Updated: Sep 28, 2009 8:35 PM⏎Source

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 of course, this is a really good book to get the best of the month. So I have been thinking about lately and the reason for this is a book that's got a lot of things. So the whole is a book that

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

    The industry's own American E-Liquid Manufacturing Standards Association requires that nicotine content be within 10 percent of the labeled content, the health department said.⏎Seventy-three of the 12

  2. #2 peak

    without the patient actually being cut open!⏎The discovery of x-rays was a happy accident. A physics professor named Wilhelm Roentgen was testing whether cathode rays could pass through glass and observed a mysterious light

  3. #3 peak

    As of May 2nd, 10 of the 22 were still in hospital and in stable condition.⏎Kalikot District, Nepal⏎Seventeen villages in this western Nepal district have been struggling with an outbreak of diarrhea,

  4. #4 peak

    69–1870)|⏎|Convoked by||Paul III|⏎about 255 during the final sessions⏎Documents and statements⏎Seventeen dogmatic decrees covering then-disputed aspects of Catholic religion⏎|Chron

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. L0:1224 suppresses
  2. L0:178 suppresses
  3. L0:1301 weakly suppresses
  4. L2:75 weakly suppresses
  5. L1:1029 weakly excites
  6. L0:644 weakly suppresses
  7. L2:1321 weakly suppresses
  8. L1:21 weakly excites
  9. L1:2922 weakly excites
  10. L1:87 weakly suppresses
  11. L1:1973 weakly suppresses
  12. L2:111 weakly suppresses
  13. L0:1405 weakly excites
  14. L2:354 weakly excites
  15. L2:41 weakly suppresses
  16. L1:1630 weakly excites
  17. L1:2801 weakly excites
  18. L1:1593 weakly suppresses
  19. L1:951 weakly excites
  20. L1:2905 weakly suppresses
  21. L2:1496 weakly suppresses
  22. L1:1857 weakly suppresses
  23. L0:1472 weakly excites
  24. L2:757 weakly suppresses

Feeds later layers

  1. L4:135 excites
  2. L4:428 suppresses
  3. L5:983 excites
  4. L4:285 excites
  5. L5:1950 excites unclear: 'Lorenz' and name fragments
  6. L5:20 excites
  7. L17:2100 weakly excites
  8. L4:945 weakly excites
  9. L5:384 weakly suppresses
  10. L16:543 weakly suppresses
  11. L6:105 weakly excites
  12. L18:372 weakly excites
  13. L16:750 weakly excites
  14. L4:2488 weakly excites
  15. L10:2240 weakly excites
  16. L4:204 weakly suppresses
  17. L19:792 weakly excites
  18. L12:2613 weakly suppresses
  19. L8:460 weakly suppresses
  20. L4:157 weakly excites
  21. L27:3039 weakly suppresses
  22. L7:1220 weakly suppresses
  23. L21:1286 weakly excites
  24. L6:530 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.

1 values between -1.06 and -1.022 values between -0.47 and -0.432 values between -0.43 and -0.406 values between -0.40 and -0.368 values between -0.36 and -0.329 values between -0.32 and -0.2826 values between -0.28 and -0.2442 values between -0.24 and -0.2063 values between -0.20 and -0.16110 values between -0.16 and -0.12189 values between -0.12 and -0.08372 values between -0.08 and -0.04745 values between -0.04 and -0.00952 values between -0.00 and 0.04624 values between 0.04 and 0.08346 values between 0.08 and 0.12179 values between 0.12 and 0.15110 values between 0.15 and 0.1951 values between 0.19 and 0.2328 values between 0.23 and 0.2722 values between 0.27 and 0.3110 values between 0.31 and 0.357 values between 0.35 and 0.391 values between 0.39 and 0.431 values between 0.43 and 0.472 values between 0.47 and 0.51offnegativepositiveoften

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. IVO
  2. ạy
  3. lope
  4. udge
  5. казать
  6. ��
  7. pectral
  8. porém
  9. atatype
  10. actices

Suppresses

  1. ngũ
  2. ful
  3. 抽查
  4. `\
  5. 级
  6. Rebellion
  7. ,'\
  8. 帘
  9. validationResult
  10. Recognition