Neuron Explorer

Qwen3-0.6B

"we" opening a text fragment

A neuron in layer 1 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 word "we" when it appears at the very start of a passage, with no earlier context. Its signal is weak and turning it up or down did not give a clear theme.

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

    we eat.⏎Fat should contribute between 20 to 35 percent of daily

  2. #2 peak

    we are concerned specifically with the mitzvot in each parasha, I shall

  3. #3 peak

    we have never had any problems with abuse among<|endoftext|>Just where in the country will you

  4. #4 peak

    we should be including in the language guide.⏎Throughout EmbraceU we have decided to

  5. #5 peak

    we speak of a polygon or topologically of a mesh. In the context of spatial

  6. #6 peak

    we require mythical creatures, such as the dragon, because of their function as symbols or

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: I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I

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

    the notion that people have gotten what they deserved. In a 2013

  2. #2 peak

    the remaining three sites, 48TE411, 48TE

  3. #3 peak

    the stream they have investigated. If you are interested in becoming a⏎TEAM volunteer and

  4. #4 peak

    theological view. Ordained in 1868, his success as 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. L0:178 suppresses
  2. L0:1224 suppresses
  3. L0:1301 suppresses
  4. L0:1 weakly suppresses
  5. L0:2415 weakly suppresses
  6. L0:1982 weakly excites
  7. L0:2639 weakly excites unclear: weak hits on 'holiday' and 'playing', negative on 'Ada-fruit'
  8. L0:2644 weakly suppresses
  9. L0:1945 weakly suppresses
  10. L0:2159 weakly suppresses
  11. L0:1780 weakly suppresses
  12. L0:732 weakly excites
  13. L0:2875 weakly suppresses
  14. L0:2238 weakly excites
  15. L0:861 weakly suppresses
  16. L0:16 weakly excites
  17. L0:116 weakly excites
  18. L0:2440 weakly suppresses
  19. L0:59 weakly excites
  20. L0:479 weakly suppresses
  21. L0:1449 weakly suppresses
  22. L0:330 weakly excites
  23. L0:2589 weakly excites
  24. L0:2972 weakly excites

Feeds later layers

  1. L2:55 suppresses
  2. L18:91 excites
  3. L27:33 excites
  4. L26:21 suppresses
  5. L27:6 suppresses
  6. L20:59 excites
  7. L17:132 excites
  8. L17:15 suppresses
  9. L2:646 suppresses
  10. L22:14 excites
  11. L25:30 weakly excites
  12. L25:17 weakly excites
  13. L16:723 weakly suppresses
  14. L26:48 weakly excites
  15. L6:2807 weakly excites
  16. L2:666 weakly suppresses
  17. L27:97 weakly excites
  18. L16:1421 weakly excites
  19. L27:85 weakly excites
  20. L25:109 weakly suppresses
  21. L27:11 weakly suppresses unclear: mixed fragments in lists and pronunciations
  22. L22:16 weakly suppresses comma in dates and birth lines
  23. L2:1489 weakly suppresses
  24. L2:321 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 -3.21 and -3.111 values between -1.07 and -0.971 values between -0.97 and -0.882 values between -0.79 and -0.691 values between -0.60 and -0.511 values between -0.51 and -0.412 values between -0.41 and -0.3218 values between -0.32 and -0.2389 values between -0.23 and -0.13467 values between -0.13 and -0.042508 values between -0.04 and 0.05648 values between 0.05 and 0.15138 values between 0.15 and 0.2423 values between 0.24 and 0.335 values between 0.33 and 0.423 values between 0.42 and 0.52offnegativepositiveoften

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. 当局
  2. icient
  3. 最后一个
  4. xmax
  5. öd
  6. 不会有
  7. edeceği
  8. apest
  9. =max
  10. .getMax

Suppresses

  1. v
  2. br
  3. def
  4. s
  5. p
  6. trade
  7. CRT
  8. com
  9. �
  10. r