Neuron Explorer

Qwen3-0.6B

the letter r starting a word piece

A neuron in layer 26 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 reacts to a single 'r' or 'A' that starts a fragment of a longer word, such as the 'r' in 'ratiocinate' or 'rancor'. When it was turned up, the model fell into repeating short phrases, with no clear topic.

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

    proportion.⏎5. (n.) To exercise the rational faculty; to deduce inferences from premises; to perform the process of deduction or of induction; to ratiocinate; to reach conclusions

  2. #2 peak

    death, but he only thought that this relationship between Romeo and Juliet be the cure to their families grudge for each other.⏎Their love is forbidden because of a rancor between their families. As they arrange a marriage behind everyone’s back,

  3. #3 peak

    by a group of nomads called the Aryans living in what is now Southern Russia. It is of course critical to mention the improper usage of the word "Aryan" in N

  4. #4 peak

    the prey, my son, you have gone up. He stooped down; he crouched as a lion and as a lioness; who dares rouse him? The<|endoftext|>Juvenile striped bass spawned in the Chesapeake Bay were

  5. #5 peak

    in the province of Taurida. He purchased sheep from the tsar's own flocks and also imported some rams from Saxony. At that time the average Mennonite farm had between

  6. #6 peak

    prolific; they breed at about six months old, producing four or five litters, of from four to ten young ones in each, in the year. The raptorial birds keep down their numbers in

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: like, like, like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like like

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

    OR REARRANGEMENT REACTION. THIS EXPERIMENT WILL DETERMINE THE FACTORS OF ELECTROPHILIC SUBSTITUTION ON DIFFERENT AROMATIC RINGS TH

  2. #2 peak

    ЛЬНАЯ РАБОТА ПО АНГЛИЙСКОМУ ЯЗЫКУ⏎КОН

  3. #3 peak

    How Humidity Affects Flu Outbreaks⏎03-05-2009

  4. #4 peak

    the most part, you will have positive outcomes. Though the process can be long, it is still worth the steps required to get greater hearing abilities.⏎Programming Hearing Aids

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. L23:2833 excites
  2. L21:1072 suppresses
  3. L17:278 excites 'as' in 'as well as' and 'as much as'
  4. L18:102 suppresses
  5. L6:328 suppresses
  6. L2:1699 excites
  7. L16:84 excites
  8. L20:174 excites
  9. L22:961 suppresses
  10. L16:628 weakly suppresses
  11. L19:171 weakly suppresses
  12. L16:8 weakly excites
  13. L0:49 weakly excites
  14. L0:2837 weakly excites
  15. L10:103 weakly excites
  16. L16:63 weakly suppresses
  17. L10:237 weakly suppresses
  18. L5:779 weakly suppresses
  19. L19:569 weakly excites
  20. L0:481 weakly suppresses
  21. L4:430 weakly suppresses
  22. L21:1233 weakly suppresses
  23. L20:601 weakly excites
  24. L20:950 weakly suppresses

Feeds later layers

  1. L27:2769 weakly excites capital 'K' starting a name piece
  2. L27:1546 weakly excites
  3. L27:1086 weakly excites
  4. L27:1082 weakly excites
  5. L27:1560 weakly excites
  6. L27:564 weakly suppresses
  7. L27:1090 weakly excites
  8. L27:793 weakly excites
  9. L27:602 weakly suppresses
  10. L27:1920 weakly excites
  11. L27:1488 weakly suppresses
  12. L27:102 weakly excites
  13. L27:1911 weakly excites
  14. L27:2490 weakly excites
  15. L27:1833 weakly suppresses
  16. L27:582 weakly suppresses
  17. L27:834 weakly excites
  18. L27:2585 weakly excites
  19. L27:408 weakly suppresses
  20. L27:2516 weakly excites
  21. L27:1006 weakly excites
  22. L27:1282 weakly suppresses
  23. L27:2502 weakly suppresses
  24. L27:1349 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.

3 values between -42.41 and -40.711 values between -40.71 and -39.011 values between -28.84 and -27.143 values between -23.75 and -22.051 values between -22.05 and -20.361 values between -18.66 and -16.973 values between -16.97 and -15.271 values between -15.27 and -13.574 values between -13.57 and -11.885 values between -11.88 and -10.1810 values between -10.18 and -8.484 values between -8.48 and -6.797 values between -6.79 and -5.0921 values between -5.09 and -3.4048 values between -3.40 and -1.701471 values between -1.70 and -0.002269 values between -0.00 and 1.6925 values between 1.69 and 3.3910 values between 3.39 and 5.086 values between 5.08 and 6.784 values between 6.78 and 8.483 values between 8.48 and 10.171 values between 10.17 and 11.871 values between 11.87 and 13.571 values between 13.57 and 15.261 values between 16.96 and 18.651 values between 20.35 and 22.051 values between 22.05 and 23.741 values between 23.74 and 25.44offnegativepositiveoften

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. лер
  2. iel
  3. réponse
  4. цию
  5. 律
  6. ту
  7. óc
  8. uitive
  9. uild
  10. раз

Suppresses

  1. 迢
  2. est
  3. curs
  4. ñe
  5. いか
  6. EST
  7. 书记
  8. borne
  9. зрения
  10. 컷