Neuron Explorer

Qwen3-0.6B

day number in dates

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 weakly on the spot where the day number goes in a date, such as 'Published on Jan 10, 2016'. It is a weak signal, and turning it up did not produce a clear theme in the written text.

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

    Quite naturally, it will take some time for these acquisitions to start making a positive impact on revenue and, by<|endoftext|>Laser Ignition System⏎Published on Jan 10, 2016⏎Economic as well

  2. #2 peak

    Digitals by Danielle Thompson Chalkboard Paper Kit. Fonts used: school script dashed, Rudiment and Homemade Apple.⏎Posted by Craftyminx on October 12, 2011 at

  3. #3 peak

    ) Limited, Delta, BC, Canada (Martyr Brand), www.canmet.com<|endoftext|>5 interesting facts about the Singapore River clean-up⏎Published on Jul 5, 2014 6:00 AM⏎The heavily

  4. #4 peak

    phosis⏎Updated: Mar 25, 2022⏎Your spine consists of a series of

  5. #5 peak

    closed.<|endoftext|>The bittersweet truth about how taste works⏎Scientists have discovered an ion channel that allows taste buds to communicate with the brain.⏎Wed, Mar 06, 2013 at 03:20 PM

  6. #6 peak

    Yes, my child is one of the 12 percent--now what are we going to do about it?"⏎How about it?<|endoftext|>Published on Feb 20, 2013⏎CBS News RAW: A new eruption

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: "The difference in the difference of two things" - the difference of two things - the difference of two things - the difference of two things - the difference in the same things - the difference in the same things - the difference in the same

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

    this topic I highly recommend reading Bruce Rogers' post:⏎Good study for the Moxy people. I did a post and review on that article. https://t.co/D9NIorgXuc— Bruce Rogers (@bjrmd) August

  2. #2 peak

    % of the stories) and Dashkat Hakaayuʔ “Whitebread” (who told 19% of the stories) https://archive.org/details/traditionsofcadd00dorsrich⏎Explore more about the

  3. #3 peak

    with each Dataforth SCM. Dataforth test instruments are calibrated and traceable to NIST standards. (see NIST info at: http://www.nist.gov/public_affairs/labs2.htm)View an example of a Dataforth test

  4. #4 peak

    so arranged that the first blow for each hour shall be accurate to the second of time." No clockmaker thought it remotely possible to<|endoftext|>by George Heymann | @te

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. L8:7 excites
  2. L8:3 weakly excites
  3. L7:2316 weakly excites
  4. L8:567 weakly excites
  5. L1:188 weakly excites
  6. L8:388 weakly suppresses
  7. L2:1250 weakly suppresses
  8. L0:28 weakly suppresses
  9. L4:56 weakly excites
  10. L8:2879 weakly suppresses
  11. L0:1718 weakly excites
  12. L1:1487 weakly suppresses
  13. L4:1408 weakly excites
  14. L3:1117 weakly suppresses
  15. L6:925 weakly suppresses
  16. L6:1520 weakly suppresses
  17. L8:4 weakly excites
  18. L5:1187 weakly excites
  19. L0:2145 weakly excites
  20. L5:922 weakly suppresses
  21. L1:1374 weakly excites
  22. L7:2831 weakly suppresses
  23. L5:2757 weakly suppresses
  24. L7:247 weakly suppresses

Feeds later layers

  1. L20:1356 weakly excites
  2. L13:2601 weakly suppresses
  3. L16:1013 weakly suppresses
  4. L16:676 weakly suppresses
  5. L15:407 weakly suppresses
  6. L22:1738 weakly excites
  7. L13:2898 weakly suppresses
  8. L15:1927 weakly excites
  9. L19:1549 weakly excites
  10. L21:1801 weakly excites
  11. L23:59 weakly suppresses
  12. L21:180 weakly suppresses
  13. L14:1864 weakly excites
  14. L11:2387 weakly suppresses
  15. L14:3017 weakly excites
  16. L11:2986 weakly excites
  17. L13:1733 weakly excites
  18. L12:1861 weakly excites
  19. L16:1406 weakly excites
  20. L14:1304 weakly excites
  21. L23:1655 weakly excites
  22. L10:1266 weakly excites
  23. L15:2399 weakly suppresses
  24. L14:2615 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 -1.54 and -1.481 values between -0.87 and -0.811 values between -0.62 and -0.561 values between -0.56 and -0.507 values between -0.50 and -0.4418 values between -0.44 and -0.3844 values between -0.38 and -0.3274 values between -0.32 and -0.26150 values between -0.26 and -0.20239 values between -0.20 and -0.14377 values between -0.14 and -0.08523 values between -0.08 and -0.01659 values between -0.01 and 0.05542 values between 0.05 and 0.11447 values between 0.11 and 0.17348 values between 0.17 and 0.23196 values between 0.23 and 0.29130 values between 0.29 and 0.3574 values between 0.35 and 0.4141 values between 0.41 and 0.4721 values between 0.47 and 0.546 values between 0.54 and 0.607 values between 0.60 and 0.661 values between 0.84 and 0.90offnegativepositiveoften

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. ote
  2. AccessException
  3. andes
  4. 能在
  5. otta
  6. acı
  7. фор
  8. over
  9. -over
  10. кон

Suppresses

  1. onica
  2. $ ⏎
  3. ruled
  4. uchos
  5. 两者
  6. aan
  7. REW
  8. ') ⏎
  9. ério
  10. _imag