Neuron Explorer

Qwen3-0.6B

unclear: maintenance, safety and process nouns

A neuron in layer 13 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on a mix of words in everyday instructions and descriptions, like 'replacing' air filters, 'leaves' the pool, or 'quickly' in a process. The pattern is loose. When turned up, the model drifted into unrelated topics like doctors, a family, and a story.

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

    boils (clean) water, producing steam, which turns a turbine and produces electricity. If the cooling process stops for any reason, or the chain reaction moves too quickly, the fuel will overheat. The fuel pellets are typically sealed inside zircon

  2. #2 peak

    visual contact. Remain close to your child when in a pool, spa or bathtub.⏎- Teach all children to get out of the water if a supervising adult leaves the pool area.⏎B = Barriers (for your pool)⏎Installation and proper

  3. #3 peak

    quickly, ensuring timely and accurate repairs or replacements.⏎The Importance of Regular Maintenance⏎Maintenance is vital for the longevity of HVAC systems. This includes routine tasks such as replacing air filters, cleaning coils, inspecting ductwork, and ensuring all moving parts

  4. #4 peak

    repairs are set right by the Laboratory assistants and teaching staff; and the major repairs are handled by the Technical Assistant and are reused. ACET has entered MoU

  5. #5 peak

    osaics.⏎The cookie settings on this website are set to "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing your cookie settings or you click "Accept" below then you are consenting to

  6. #6 peak

    of the tears, allowing them to stay on the eye surface longer. The doctors can insert these plugs in a matter of minutes, during an office visit. After placement, the plugs cannot be seen, won’t rub out or scratch your cornea

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: This is a PPT that I have a PPT: I need to do a PPT that I have a CHINE! A. A B C D A. This is a PPT that I have a J I need a

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

    in 1769, and this number increased to 7,600 in 1801.Aker was established as a municipality January 1, 1838 (see formannskapsdistrikt

  2. #2 peak

    With Precautions⏎The agency recommends thoroughly cleaning feeders at least once a week with a bleach solution and to take down your feeder for a week if you notice sick or dead birds.⏎Officials still haven't figured out the cause of the '

  3. #3 peak

    in your professional development<|endoftext|>Mackinder’s Heartland Theory made SIMPLE⏎Do you want to learn about Mackinder’s Heartland Theory? Then you have come to the right

  4. #4 peak

    . Theoretical modeling specifically points to quasar outflows as the culprit. In this negative feedback loop, while the black hole is actively acquiring mass as a quasar, the outflows carry away energy and material, suppressing further growth. Small-scale

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. L11:2039 excites
  2. L11:993 suppresses
  3. L11:1745 suppresses
  4. L11:2418 suppresses
  5. L12:2568 suppresses
  6. L11:571 suppresses
  7. L11:1349 excites
  8. L12:1436 suppresses
  9. L10:1222 suppresses
  10. L12:2156 suppresses
  11. L12:1858 suppresses
  12. L10:1374 suppresses
  13. L9:1336 suppresses
  14. L11:121 suppresses
  15. L10:2852 suppresses
  16. L12:2148 suppresses
  17. L11:1310 suppresses
  18. L12:2117 suppresses
  19. L12:1183 suppresses
  20. L11:1752 excites
  21. L11:743 excites
  22. L10:2177 excites
  23. L11:2189 weakly excites
  24. L11:1854 weakly suppresses

Feeds later layers

  1. L14:1112 suppresses
  2. L14:715 suppresses
  3. L14:792 excites unclear: mid-phrase words before a noun
  4. L19:457 suppresses
  5. L14:2516 suppresses
  6. L14:2750 suppresses
  7. L21:1453 suppresses
  8. L16:1031 suppresses
  9. L14:2979 suppresses
  10. L14:2297 suppresses
  11. L16:2441 suppresses
  12. L14:249 excites
  13. L14:2267 excites
  14. L15:530 suppresses
  15. L18:2416 suppresses
  16. L15:2503 suppresses
  17. L20:2299 suppresses
  18. L14:1552 excites
  19. L17:1942 suppresses
  20. L18:2274 suppresses
  21. L21:1231 weakly suppresses
  22. L15:2114 weakly suppresses
  23. L14:754 weakly suppresses
  24. L14:324 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 -2.09 and -2.011 values between -1.85 and -1.771 values between -1.60 and -1.523 values between -1.52 and -1.442 values between -1.44 and -1.361 values between -1.36 and -1.273 values between -1.27 and -1.194 values between -1.19 and -1.114 values between -1.11 and -1.038 values between -1.03 and -0.956 values between -0.95 and -0.8617 values between -0.86 and -0.789 values between -0.78 and -0.7016 values between -0.70 and -0.6238 values between -0.62 and -0.5459 values between -0.54 and -0.4671 values between -0.46 and -0.37106 values between -0.37 and -0.29193 values between -0.29 and -0.21283 values between -0.21 and -0.13606 values between -0.13 and -0.051553 values between -0.05 and 0.04501 values between 0.04 and 0.12197 values between 0.12 and 0.2086 values between 0.20 and 0.2844 values between 0.28 and 0.3745 values between 0.37 and 0.4518 values between 0.45 and 0.538 values between 0.53 and 0.617 values between 0.61 and 0.694 values between 0.69 and 0.787 values between 0.78 and 0.863 values between 0.86 and 0.941 values between 1.02 and 1.102 values between 1.10 and 1.19offnegativepositiveoften

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. ALES
  2. ��
  3. newPath
  4. afil
  5. TypeDef
  6. ales
  7. みたいです
  8. .NewRequest
  9. rett
  10. AsStringAsync

Suppresses

  1. déjà
  2. already
  3. 早已
  4. -Con
  5. 已经是
  6. lying
  7. 已有
  8. 现有的
  9. ika
  10. already