Neuron Explorer

Qwen3-0.6B

mid-word fragments in technical terms

A neuron in layer 14 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on a piece in the middle of a longer word that has been split up, like the 'ump' in 'circumpolar' or the 'ich' in 'Omnichord'. Turning it up did not produce any clear change in topic.

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

    central to north western Canada. There are four subspecies described by Hall (1981) with two (M.v.interior and M.v. longicrus) occurring in California.⏎This is a common bat found in all the mountain

  2. #2 peak

    OK state<|endoftext|>The Status of Marine Mammals⏎A new multinational study, the first global review of Arctic marine mammals, has assessed the status of all circumpolar species and subpopulations, and highlights the precarious state these animals are in

  3. #3 peak

    the future of marine mammals in the region at risk. A new multinational study, the first global review of Arctic marine mammals, has assessed the status of all circumpolar species and subpopulations, and highlights the precarious state these animals are in

  4. #4 peak

    region adjacent to Terre Adélie and George V Land.Altogether, CAML revealed that Antarctica is a single bioregion united by the Circumpolar Current – the largest current in the world. Previously thought to be low in

  5. #5 peak

    zinc and other metals such as lead and zinc or even a composite plated steel plate. This steel plate not only has excellent antirust performance but also has good coating performance.<|endoftext|>- Groundbreaking held for Museum

  6. #6 peak

    ing, harplike timbre and its value as a kitsch object.⏎Suzuki introduced the Omnichord along with the Tronichord, renamed the Portachord on some

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: you can think of it as a graph, and you can think of it as this... [insert] [ ] This [ ] can be here [ ] [ ] ? ? [ ] ? This is [ ] ?

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

    00 Hz) are naturally generated by skeletal muscles during voluntary contractions. Recording of these vibrations at the muscle surface are called surface mechanomyograms (S-MMGs). In this study, S-MMGs were recorded over a 3 x

  2. #2 peak

    Death Penalty" Washington Law Review (2010).<|endoftext|>Last updated 2/27/15⏎Isothermal Community College⏎P.O. Box 804, Spindale NC 28160⏎

  3. #3 peak

    , the results obtained may also lead researchers to perform similar studies in more clinically relevant models. Proceedings of the National Academy of Sciences USA (PNAS) has published the study online this week.⏎Argentinian scientist André

  4. #4 peak

    ankovic J, Mazziotta JC, eds. Bradley's Neurology in Clinical Practice. 6th ed. Philadelphia, PA: Elsevier Saunders; 2012:chap 68.⏎G

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. L10:709 strongly suppresses
  2. L7:943 strongly suppresses
  3. L8:1640 strongly suppresses
  4. L12:2736 suppresses
  5. L4:2029 excites
  6. L4:1854 excites
  7. L9:1777 suppresses
  8. L8:1699 suppresses
  9. L6:3059 excites
  10. L4:370 suppresses
  11. L5:807 excites
  12. L10:1028 suppresses
  13. L4:2830 excites
  14. L6:2400 suppresses
  15. L4:2360 excites
  16. L10:1961 excites
  17. L13:874 excites
  18. L5:1273 excites
  19. L7:210 suppresses
  20. L3:494 suppresses
  21. L13:2410 excites
  22. L4:1872 suppresses
  23. L5:1053 suppresses
  24. L4:1088 suppresses

Feeds later layers

  1. L18:1116 strongly excites
  2. L17:1990 excites
  3. L21:746 excites
  4. L17:1089 excites
  5. L20:838 excites
  6. L16:104 excites
  7. L19:614 excites
  8. L22:164 excites
  9. L18:170 excites
  10. L25:578 excites
  11. L17:152 excites mid-word fragments, often 'intric'
  12. L19:672 excites
  13. L21:747 excites
  14. L16:582 excites
  15. L15:2427 excites
  16. L17:45 suppresses
  17. L16:415 excites
  18. L21:443 excites
  19. L19:1747 excites
  20. L16:286 excites
  21. L16:1590 excites
  22. L15:1900 suppresses
  23. L16:2720 excites
  24. L19:2386 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.

2 values between -1.06 and -0.8028 values between -0.80 and -0.54347 values between -0.54 and -0.282121 values between -0.28 and -0.031301 values between -0.03 and 0.2353 values between 0.23 and 0.4914 values between 0.49 and 0.757 values between 0.75 and 1.003 values between 1.00 and 1.264 values between 1.26 and 1.524 values between 1.52 and 1.782 values between 1.78 and 2.031 values between 2.03 and 2.291 values between 2.29 and 2.553 values between 2.55 and 2.811 values between 3.06 and 3.323 values between 3.32 and 3.581 values between 3.58 and 3.842 values between 3.84 and 4.091 values between 4.09 and 4.351 values between 4.35 and 4.611 values between 4.61 and 4.871 values between 5.12 and 5.382 values between 6.67 and 6.932 values between 6.93 and 7.182 values between 8.99 and 9.24offnegativepositiveoften

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. ัน
  2. awn
  3. ffen
  4. grids
  5. ubbo
  6. od
  7. ognito
  8. sake
  9. oods
  10. evin

Suppresses

  1. 点击查看
  2. 但她
  3. sez
  4. BIT
  5. zeitig
  6. particul
  7. 抬头
  8. ße
  9. weis
  10. weg