Neuron Explorer

Qwen3-0.6B

just before a document ends

A neuron in layer 21 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 last word of a web page or article, right before the text break that marks the end of one document and the start of another. Turning it up produced rambling, repetitive text rather than anything about endings.

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

    220.127.116.11, you can define a List of Users and Cities. Create a new variable and:⏎Just<|endoftext|>George Washington was faced with many challenges during his presidency as he not only had to

  2. #2 peak

    State, and Cornie Scheffer, professor of biomedical engineering at Stellenbosch University, served as mentors on this project.⏎A report on Penn State Live<|endoftext|>You might have seen a vaping advert, whether you are a parent with a child

  3. #3 peak

    Chile to speed up its judicial process and create a national plan to search for disappeared persons.⏎From Jurist, Aug. 23. Used with permission.<|endoftext|>Oise River, river that rises in Belgium in the Ardennes mountains, southeast

  4. #4 peak

    These snakes are harmless to humans and are helpful to keep rodent populations in check. If you see one, just let them be and wave from a distance.<|endoftext|>How to Read Diode Numbers⏎By James Miller⏎The system for categorizing

  5. #5 peak

    (picking which cells are initially turned on) and watching the world behave accordingly to the set of rules.⏎View/Play Conway's Game of Life now!<|endoftext|>It may still be difficult for some people to accept that an activity such as gambling

  6. #6 peak

    9", "crochet school", "learn to crochet", "treble crochet", crochet⏎| Digg This⏎| Save to del.icio.us<|endoftext|>Back in the early 20th century, the mountains and hollers

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: how to make a subject to the right size to the light in the dark to the environment to the health to the environment to the health to the environment to the health to the environment to the health to the environment to the health to the environment

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

    only consumes constant space.⏎In an imperative language (here Modula-3) the same would be achieved with the following code:⏎TYPE List = REF RECORD next: List; value: INTEGER; END;⏎VAR x := NEW(List, value:=

  2. #2 peak

    and Physiology in Health and Illness 12e⏎Goodreads helps you keep track of books you want to read. Want to Read saving…. Want to Read Currently Reading Read. Other editions. Enlarge cover. Error rating book.⏎Browse

  3. #3 peak

    C and Immune Function, 2017, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5707683/#:~:text=It%20is%20a%20

  4. #4 peak

    ATRIP complex induces the Fanconi Anemia repair and Nucleotide Excision Repair (NER) pathways, as well as the p53 signaling pathway. C. Finally, a

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. L5:601 strongly suppresses
  2. L6:449 strongly excites
  3. L16:37 strongly excites
  4. L18:555 strongly suppresses end-of-text marker between documents
  5. L18:221 suppresses
  6. L0:54 weakly suppresses
  7. L19:197 weakly suppresses
  8. L19:798 weakly suppresses
  9. L4:2339 weakly suppresses
  10. L20:299 weakly suppresses
  11. L19:283 weakly suppresses
  12. L3:50 weakly excites
  13. L20:1398 weakly suppresses
  14. L4:97 weakly suppresses
  15. L20:249 weakly suppresses
  16. L20:221 weakly suppresses
  17. L10:3 weakly excites
  18. L18:931 weakly excites
  19. L20:2371 weakly excites
  20. L5:2458 weakly excites
  21. L4:804 weakly suppresses
  22. L16:454 weakly suppresses
  23. L1:141 weakly excites
  24. L16:93 weakly suppresses

Feeds later layers

  1. L25:436 suppresses
  2. L27:1300 suppresses
  3. L22:230 suppresses
  4. L22:133 weakly excites
  5. L26:1349 weakly suppresses
  6. L27:158 weakly excites
  7. L23:2081 weakly suppresses
  8. L27:10 weakly excites
  9. L26:2922 weakly suppresses
  10. L26:1356 weakly excites
  11. L27:1962 weakly suppresses
  12. L26:2341 weakly excites
  13. L23:2483 weakly excites
  14. L24:264 weakly excites
  15. L27:2050 weakly suppresses
  16. L27:2818 weakly suppresses
  17. L25:2777 weakly suppresses
  18. L25:939 weakly suppresses
  19. L23:2142 weakly excites
  20. L22:2283 weakly suppresses
  21. L25:827 weakly suppresses
  22. L22:2324 weakly suppresses
  23. L25:2135 weakly excites
  24. L22:300 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 -6.06 and -5.312 values between -3.79 and -3.035 values between -2.28 and -1.5224 values between -1.52 and -0.762360 values between -0.76 and -0.011438 values between -0.01 and 0.7516 values between 0.75 and 1.5115 values between 1.51 and 2.2710 values between 2.27 and 3.028 values between 3.02 and 3.785 values between 3.78 and 4.547 values between 4.54 and 5.292 values between 5.29 and 6.052 values between 6.05 and 6.812 values between 6.81 and 7.561 values between 7.56 and 8.321 values between 8.32 and 9.081 values between 9.08 and 9.841 values between 9.84 and 10.592 values between 11.35 and 12.111 values between 14.38 and 15.131 values between 15.13 and 15.891 values between 15.89 and 16.651 values between 17.41 and 18.161 values between 23.46 and 24.22offnegativepositiveoften

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. UNCH
  2. zap
  3. van
  4. amb
  5. d
  6. files
  7. endet
  8. ucumber
  9. -files
  10. van

Suppresses

  1. 干什么
  2. ares
  3. iais
  4. theless
  5. asily
  6. 的主题
  7. 勾
  8. uble
  9. 徂
  10. semiclass