Neuron Explorer

Qwen3-0.6B

'in' after 'set' or 'kick'

A neuron in layer 3 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 lights up on the word 'in' when it finishes phrases like 'set in' or 'kicks in', meaning something starts to take hold. When it was turned up, the model's replies did not clearly show this theme.

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

    risen astronomically. Bayer too was affected by these developments. The crisis reached its apex in the early 1980s as a severe global recession set in.⏎Expansion of foreign business⏎Despite the unfavorable environment, Bayer expanded its international activities

  2. #2 peak

    ed that religion has made ample room for science over the⏎The question<|endoftext|>On June 28, 2013, a rogue lightning strike set in motion a series of events near Yarnell, Arizona, that would ultimately end

  3. #3 peak

    in severity, but generally begin with memory issues. Over time, as forgetfulness and confusion set in, personal hygiene can suffer and social interactions become challenging. This can ultimately lead to

  4. #4 peak

    , short temper and anger.⏎PARASYMPATHETIC NERVOUS SYSTEM⏎Once the threat is over, our parasympathetic mode kicks in<|endoftext|>The use of mathematical modeling in engineering allows for a significant reduction of material costs

  5. #5 peak

    to make sure that we don’t forget it. Yet we seemingly find no such reminder … or do we?⏎The truth is that the Rabbis did set in place a ritual in the synagogue which serves to remind us of the Revelation at Sinai

  6. #6 peak

    or improper grooming and trimming of the nail.⏎Ingrown toenails should be treated as soon as they are recognized. If they are recognized early (before infection sets in), home care may prevent the need for further treatment:⏎- Soak the foot

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 am a proponent that the number of the day of (x) is not to say that. But that it is... the number of the day. The number of the... the number... the number of the... the number of

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

    my eyes in the darkness, I struggled to find the source of the commotion. Through my binoculars, I could just a make out the silhouette of a subadult grizzly devouring salmon carcasses on the river

  2. #2 peak

    with regarding one's career as a deeply serious matter too and is a prime source of humour on The Apprentice (the British version at least) - contestants who make out that their unremarkable junior management positions were the equivalent of founding Google or Microsoft

  3. #3 peak

    7, 1862.⏎This navy blue infantry battle regalia is made out of wool according to Union color standards of the era. It sports double breasted

  4. #4 peak

    dissolved carbon dioxide and sunlight energy. The carbon dioxide is dispersed into the reactor fluid to make it accessible for the algae. The bioreactor has to be made out of transparent material.⏎The algae are photoautotroph organisms which perform oxygenic

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. L2:2190 excites
  2. L2:70 weakly excites
  3. L0:1480 weakly suppresses
  4. L2:382 weakly suppresses
  5. L1:951 weakly excites
  6. L2:2395 weakly excites
  7. L0:1436 weakly excites
  8. L2:1657 weakly excites
  9. L0:2855 weakly suppresses
  10. L2:493 weakly suppresses
  11. L0:1576 weakly suppresses
  12. L1:1539 weakly suppresses
  13. L0:1081 weakly suppresses
  14. L1:2322 weakly suppresses
  15. L1:2138 weakly suppresses
  16. L2:1710 weakly suppresses
  17. L0:66 weakly suppresses
  18. L1:1591 weakly excites
  19. L1:708 weakly suppresses end-of-document marker token
  20. L1:2353 weakly suppresses
  21. L1:2441 weakly suppresses
  22. L2:237 weakly excites
  23. L2:1321 weakly suppresses
  24. L2:2566 weakly excites

Feeds later layers

  1. L5:2799 suppresses
  2. L4:1397 suppresses
  3. L10:1167 weakly excites
  4. L4:607 weakly excites
  5. L4:2795 weakly excites
  6. L16:2954 weakly excites
  7. L6:2656 weakly suppresses
  8. L6:2335 weakly suppresses
  9. L6:1742 weakly excites
  10. L11:329 weakly suppresses number or citation fragment split
  11. L14:1344 weakly excites
  12. L8:1007 weakly suppresses
  13. L9:2534 weakly suppresses
  14. L27:1837 weakly excites
  15. L14:2101 weakly excites
  16. L7:395 weakly excites
  17. L20:2735 weakly excites
  18. L18:2826 weakly suppresses
  19. L18:2716 weakly suppresses
  20. L14:402 weakly suppresses
  21. L17:1440 weakly suppresses
  22. L25:697 weakly suppresses
  23. L9:987 weakly excites
  24. L18:2655 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.

2 values between -2.49 and -2.351 values between -1.61 and -1.461 values between -1.31 and -1.173 values between -0.87 and -0.721 values between -0.72 and -0.584 values between -0.58 and -0.4332 values between -0.43 and -0.28270 values between -0.28 and -0.132017 values between -0.13 and 0.021362 values between 0.02 and 0.16191 values between 0.16 and 0.3118 values between 0.31 and 0.464 values between 0.46 and 0.611 values between 1.64 and 1.791 values between 3.26 and 3.41offnegativepositiveoften

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. IOR
  2. uhe
  3. like
  4. InstanceId
  5. núi
  6. aired
  7. 垤
  8. ↩
  9. yours
  10. 因为你

Suppresses

  1. (dAtA
  2. väl
  3. femin
  4. templates
  5. obuf
  6. .setSelection
  7. 他们都
  8. agreed
  9. sequelize
  10. 下乡