Neuron Explorer

Qwen3-0.6B

"and" after a contrast or pause

A neuron in layer 6 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 "and" when it starts a new main clause after a comma, semicolon or a word like "however", as in "the results are surprising, and will provoke debate". Steering did not produce any clear related writing.

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

    had attended a Gael-Linn language enrichment course, but the rest had little or no exposure to Irish. The results of this research are both striking and surprising, and will provoke fresh debate on the role of the Irish language in Northern Ireland today.

  2. #2 peak

    . He was tall, erect, and of fine physique. For a few moments they carried on what appeared to be a friendly conversation, much to our relief; and then the entire group started down the road toward us.⏎It was not difficult to

  3. #3 peak

    the reef to establish a coaling station while avoiding the high-taxes imposed at ports controlled by the Hawaiians. The project was an utter failure, however, and while evacuating the last of the workers, the USS Saginaw ran ag

  4. #4 peak

    on or transgene insertions. This is not believed to be a widespread phenomenon in mammals, however, and you don’t need to (and shouldn’t!) take my word for it –

  5. #5 peak

    had captured from the Turks, ordered up General Cathcart’s brigade to prevent them from effecting their object. Some delay occurred before the brigade began its march; and the Commander-in-Chief, seeing that the Russians would succeed in carrying off the

  6. #6 peak

    , Indians killed a third of the one thousand settlers in Virginia. Another serious attack came in 1644. The English colonies grew rapidly, however, and after about 1650 had much larger adult male populations than the Indians

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: "10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 101, 1

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

    work on the structure of the virus shells and identification of mutations to improve their stability was carried out by Professor David Stuart and his team at Oxford University using Diamond Light Source, the UK's national<|endoftext|>Coronavirus (COVID-19) and

  2. #2 peak

    and life science director at Diamond Light Source, and Dr Bryan Charleston of The Pirbright Institute. The findings are published in

  3. #3 peak

    use of epoxy-based resins and may be classified as the thermosetting epoxide polymer. Exteriors that receive this coating are resistant to ultra-v

  4. #4 peak

    y pen, or this mini plasma cutter.⏎Continue reading “Hacked CCFL Inverter Becomes An Arc Lighter

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. L4:57 suppresses
  2. L5:548 excites
  3. L4:1529 excites
  4. L3:2565 weakly excites
  5. L4:2601 weakly excites
  6. L5:131 weakly suppresses
  7. L3:652 weakly suppresses
  8. L4:1927 weakly suppresses
  9. L4:131 weakly suppresses
  10. L2:1008 weakly suppresses
  11. L3:2453 weakly excites
  12. L2:1354 weakly excites
  13. L3:229 weakly excites
  14. L5:1522 weakly suppresses
  15. L1:1250 weakly suppresses
  16. L0:1397 weakly excites
  17. L2:3006 weakly excites
  18. L5:2131 weakly excites
  19. L1:2160 weakly suppresses
  20. L4:2996 weakly excites
  21. L3:1194 weakly excites
  22. L5:1653 weakly excites
  23. L0:19 weakly excites
  24. L1:2577 weakly excites

Feeds later layers

  1. L8:410 excites
  2. L16:1112 excites
  3. L7:93 excites
  4. L16:797 suppresses
  5. L9:103 excites "as" or "since" giving a reason
  6. L7:751 suppresses
  7. L7:377 excites
  8. L17:839 excites
  9. L8:361 excites
  10. L19:2073 weakly excites
  11. L9:161 weakly excites
  12. L7:259 weakly suppresses
  13. L7:337 weakly excites
  14. L16:949 weakly excites
  15. L24:2486 weakly excites
  16. L9:2302 weakly excites
  17. L18:229 weakly suppresses
  18. L8:3048 weakly excites
  19. L20:122 weakly suppresses
  20. L18:867 weakly excites
  21. L10:1449 weakly excites
  22. L12:1285 weakly excites
  23. L12:1200 weakly excites
  24. L23:373 weakly suppresses 'to' after a noun or adjective

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 -1.45 and -1.301 values between -1.30 and -1.151 values between -1.00 and -0.855 values between -0.85 and -0.7011 values between -0.70 and -0.5524 values between -0.55 and -0.40204 values between -0.40 and -0.25873 values between -0.25 and -0.101847 values between -0.10 and 0.06733 values between 0.06 and 0.21120 values between 0.21 and 0.3625 values between 0.36 and 0.5112 values between 0.51 and 0.668 values between 0.66 and 0.818 values between 0.81 and 0.962 values between 0.96 and 1.113 values between 1.11 and 1.268 values between 1.26 and 1.414 values between 1.41 and 1.561 values between 1.56 and 1.715 values between 1.71 and 1.861 values between 1.86 and 2.011 values between 2.01 and 2.162 values between 2.16 and 2.312 values between 2.31 and 2.461 values between 2.46 and 2.611 values between 2.61 and 2.762 values between 2.91 and 3.061 values between 3.36 and 3.511 values between 4.41 and 4.56offnegativepositiveoften

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. acja
  2. 糜
  3. MLS
  4. amsung
  5. 它可以
  6. ilst
  7. 秣
  8. 到这里
  9. ,it
  10. 这是一种

Suppresses

  1. ObjectId
  2. [date
  3. Confeder
  4. اه
  5. degree
  6. 限度
  7. 厚
  8. 施行
  9. cido
  10. adj