Neuron Explorer

Qwen3-0.6B

end-of-text token and answer-stating phrases

A neuron in layer 15 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 is weak and mixed. Its strongest hits are the invisible end-of-document marker right after a finished article, and it dips on phrases like 'the answer is'. Turning it up or down gave text that varied a lot, with no clear shared 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

    Anger can surface at any time. When you feel pain, you may want to lash out at the people around you.<|endoftext|>In Northern Ireland the Irish language has the power to enrage and enthral.

  2. #2 peak

    of all bourgeois states, France and Britain prominent among them.⏎What made you want to look up "Constantinople Agreement"? Please share what surprised you most...<|endoftext|>Your healthcare provider will ask about your medical and family history and your overall health.

  3. #3 peak

    vf = vi + at⏎= 56.3m/s + (-9.81m/s2)(4.52s)⏎vf = 12.0 m/s⏎This value is still positive, but smaller.

  4. #4 peak

    Find out more information here⏎A study has revealed one in five children in Herefordshire are not fully immunised against measles. Click here for more information.<|endoftext|>What is the Raspberry Pi?⏎The Raspberry Pi is a miniature computer that is ideal

  5. #5 peak

    least 12 missions to carry cargo to and from the International Space Station as part of the Commercial Resupply Services (CRS) contract for NASA.⏎For<|endoftext|>|1st person||me me||nos us|⏎|2nd person||

  6. #6 peak

    , it’d be impossible to definitely pinpoint which aircraft faired better in the Luftwaffe. When flown against Allied bomber raids, neither was a welcome sight.<|endoftext|>Cancer Prevention and Early Detection⏎Choosing a healthy lifestyle can

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 think the answer is... and the answer is... and the answer is... and the answer is... and the answer is... and the answer is... and the answer is... and the answer is... and the answer is... and the

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

    . By present, crowded standards, almost all outdoor sports require an exorbitant amount<|endoftext|>Guess Who? Identity theft?⏎“Does your person have blue eyes?” “Does your person have massive eyebrows?” “Is your person wearing a hat?” “

  2. #2 peak

    not have 15m customers or measure their Twitter output in terabytes, so is big data truly going to have an impact on them? The short answer is yes – a significant impact – and one that is already attracting the interest of new

  3. #3 peak

    203 deaths in 2019. “How do those rates compare to the death rates from the coronavirus vaccine?” Carlson asks. The answer is, there’s really no comparison.<|endoftext|>Technology can help fight cheating that is itself

  4. #4 peak

    happens that Easter falls on the Sunday during Passover. Presents are not exchanged on Passover.⏎Question: Is Passover the first Jewish holiday the Jews celebrated?⏎Answer: You are correct -- Passover was the first holiday that the Israelites celebrated

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. L12:1263 strongly suppresses
  2. L12:1478 suppresses
  3. L12:2824 excites
  4. L12:3005 suppresses
  5. L13:1366 suppresses
  6. L11:2977 suppresses
  7. L13:1489 excites
  8. L12:1170 excites
  9. L13:2485 suppresses
  10. L13:89 suppresses
  11. L14:3008 excites
  12. L12:298 suppresses
  13. L13:1049 excites
  14. L14:1290 excites
  15. L11:1653 suppresses
  16. L14:2212 excites
  17. L11:770 excites
  18. L14:1802 weakly excites
  19. L11:324 weakly suppresses
  20. L7:2313 weakly excites "to" after "proved" or "proves"
  21. L12:2501 weakly suppresses
  22. L12:2574 weakly suppresses
  23. L13:1169 weakly excites
  24. L14:370 weakly suppresses

Feeds later layers

  1. L16:2837 suppresses
  2. L16:814 excites
  3. L16:2204 suppresses
  4. L16:1564 suppresses
  5. L16:2912 suppresses
  6. L17:867 suppresses unclear: punctuation before an explanation
  7. L16:2870 suppresses
  8. L17:1164 excites
  9. L16:1193 suppresses
  10. L19:2288 suppresses
  11. L16:1499 suppresses
  12. L17:820 suppresses
  13. L18:1544 suppresses
  14. L19:1597 suppresses
  15. L20:796 weakly suppresses
  16. L16:2370 weakly suppresses
  17. L16:1008 weakly excites
  18. L17:2207 weakly excites
  19. L19:2740 weakly suppresses
  20. L18:2739 weakly suppresses
  21. L16:2506 weakly excites
  22. L16:349 weakly suppresses
  23. L16:1067 weakly suppresses
  24. L16:347 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 -6.51 and -6.311 values between -2.90 and -2.701 values between -1.89 and -1.696 values between -1.29 and -1.097 values between -1.09 and -0.8910 values between -0.89 and -0.6912 values between -0.69 and -0.4987 values between -0.49 and -0.29749 values between -0.29 and -0.082287 values between -0.08 and 0.12598 values between 0.12 and 0.32122 values between 0.32 and 0.5222 values between 0.52 and 0.723 values between 0.72 and 0.922 values between 1.32 and 1.52offnegativepositiveoften

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. 就够了
  2. йте
  3. multipart
  4. 体重
  5. 就行了
  6. 算了
  7. dea
  8. %m
  9. (Tag
  10. cia

Suppresses

  1. Brussels
  2. Kremlin
  3. ガン
  4. aber
  5. istle
  6. rule
  7. 一律
  8. ativ
  9. สวน
  10. answer