Neuron Explorer

Qwen3-0.6B

mid-word fragments before a completion

A neuron in layer 26 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 short pieces of a longer word, such as "ad" in Benkhadra, "st" in diethylstilbestrol, or "val" in valentines, where the word is not yet finished. It goes negative on words like "brain" in puzzle-website text. Turning it up gave repetitive phrases about "the first time" and "the second half", which does not match the trigger.

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

    roles in establishing a green economy. From Secretary of State Hilary Clinton, who created a new office for women at the State Department, to Amina Benkhadra, Morocco’s Minister of Energy, Mines, Water and Environment, to Christian

  2. #2 peak

    sky changes through the seasons. We’ll hear some of the stories our ancestors told about the cosmos and how these relate to what we know today.⏎In association with STFC Rutherford Appleton Laboratory⏎If your child has additional needs, please get

  3. #3 peak

    ens experienced in the womb can lead⏎to enlarged and hypersensitive prostate in adulthood.⏎with the best known of the endocrine disrupters, particularly⏎diethylstilbestrol, or DES, led to great scientific confidence⏎about the predictive

  4. #4 peak

    network can behave chaotically during the learning phase due to large step sizes.⏎- Scott E. Fahlman: An Empirical Study of Learning Speed in Back-Propagation Networks, September 1988<|endoftext|>Cardio and weight

  5. #5 peak

    IL at George Street, a home where teen boys learn independent living skills; a community-based Diagnostic Evaluation service that helps people of all ages; SouthEast Treatment and Reentry Team (START), a program that works with teen girls in residential placement through

  6. #6 peak

    2.16 g of myristic acid, 93.96 g of palmitic acid, 9.288 g of stearic acid, 79.92 g of total monounsaturated

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: these things around public learning communities around the world classeylylylylylylylylylylylylylylylylylylylylylylylylylylylylylylylylylylylylyly

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

    ? ? spring ? ? ? hole⏎Answerwaist coat tail gate way side arm pit fall out put down cast off spring time line man hole⏎See another brain teaser just like this one...⏎Or, just get a random brain teaser⏎If

  2. #2 peak

    a matrix⏎Find nth maximum⏎Matlab Basics II - Create a vector with a repeated entry⏎Create sequnce

  3. #3 peak

    iated. Note that the spelling doesn't necessarily stay the same.⏎See another brain teaser just like this one...⏎Or, just get a random brain teaser⏎If

  4. #4 peak

    x * (x + x) * x.⏎Mile to Kilometer⏎Given A4 sizes find A3's long side⏎Polygon Interior Angle Sum⏎Bit to Gigabyte⏎Find the treasures in MATLAB Central and discover how the community can

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. L20:429 excites
  2. L5:2 weakly suppresses
  3. L16:507 weakly excites
  4. L17:123 weakly excites
  5. L16:441 weakly excites
  6. L17:366 weakly excites
  7. L25:206 weakly excites
  8. L21:355 weakly suppresses
  9. L1:2442 weakly suppresses
  10. L22:156 weakly excites
  11. L11:2505 weakly excites
  12. L14:409 weakly suppresses
  13. L9:18 weakly suppresses text boundary after a document break
  14. L25:2407 weakly excites
  15. L10:905 weakly excites
  16. L5:2683 weakly suppresses
  17. L20:1779 weakly excites
  18. L19:229 weakly excites
  19. L19:314 weakly excites
  20. L14:892 weakly suppresses
  21. L2:1086 weakly excites
  22. L7:3 weakly suppresses
  23. L2:1700 weakly excites
  24. L2:9 weakly suppresses sentence-final periods vs dots in code

Feeds later layers

  1. L27:2775 weakly suppresses
  2. L27:2765 weakly excites
  3. L27:1021 weakly suppresses
  4. L27:294 weakly suppresses
  5. L27:2317 weakly excites
  6. L27:2951 weakly excites
  7. L27:2604 weakly suppresses
  8. L27:2661 weakly suppresses
  9. L27:893 weakly suppresses
  10. L27:3016 weakly suppresses
  11. L27:332 weakly suppresses
  12. L27:1968 weakly suppresses
  13. L27:2152 weakly excites
  14. L27:1721 weakly suppresses
  15. L27:417 weakly excites
  16. L27:2615 weakly excites
  17. L27:2882 weakly excites
  18. L27:3057 weakly suppresses
  19. L27:2680 weakly suppresses
  20. L27:1908 weakly excites the 'd' that starts 'die' or 'dead'
  21. L27:2730 weakly excites
  22. L27:2327 weakly suppresses
  23. L27:2201 weakly suppresses
  24. L27:934 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 -26.88 and -25.311 values between -19.02 and -17.452 values between -17.45 and -15.882 values between -14.31 and -12.743 values between -12.74 and -11.175 values between -11.17 and -9.606 values between -9.60 and -8.0316 values between -8.03 and -6.4625 values between -6.46 and -4.8932 values between -4.89 and -3.3286 values between -3.32 and -1.75982 values between -1.75 and -0.182646 values between -0.18 and 1.3957 values between 1.39 and 2.9621 values between 2.96 and 4.5311 values between 4.53 and 6.101 values between 6.10 and 7.675 values between 7.67 and 9.242 values between 9.24 and 10.811 values between 12.38 and 13.951 values between 18.66 and 20.231 values between 20.23 and 21.811 values between 34.37 and 35.94offnegativepositiveoften

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. onenumber
  2. äd
  3. ตล
  4. าม
  5. 首发
  6. 군
  7. cycles
  8. 维奇
  9. COLUMN
  10. heets

Suppresses

  1. tool
  2. 工具
  3. Tool
  4. Tool
  5. ulaire
  6. mtx
  7. ey
  8. duel
  9. bout
  10. beiten