Neuron Explorer

Qwen3-0.6B

unclear: mixed word-start fragments

A neuron in layer 26 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

It fires on a scattering of word beginnings and sounds, such as the 'N' in 'Nausea' or 'ren' in 'renounced', with no clear shared theme. Turning it down made the model get stuck repeating 'light', and turning it up gave ordinary text, so its role is hard to pin down.

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

    agus and out of the mouth.⏎Emesis; Vomiting; Stomach upset; Upset stomach⏎Many common problems may cause nausea and vomiting:⏎Nausea and vomiting may also be early warning signs

  2. #2 peak

    security on farms⏎Biosecurity is about reducing the risk of disease transmission. We know that bTB can be transmitted by direct contact (e.g. nose-to-nose) or indirect contact (e.g. via contaminated pasture) but simple exclusion

  3. #3 peak

    - Sweating, clammy skin⏎- Enlarged (dilated) pupils⏎- Insomnia (sleeping difficulty)⏎- Loss of appetite⏎- Nausea and vomiting

  4. #4 peak

    treated thus he renounced all pardon, as he had never thought to have offended, and only asked

  5. #5 peak

    abdomen, including the stomach. This can lead to a feeling of fullness, even if you haven't eaten much, and can make you lose your appetite.⏎Nausea is a feeling of un

  6. #6 peak

    they mistake bright-colored pieces for food. Sea turtles and migrating birds can become entangled in abandoned plastic fishing nets known as “ghost nets.” Plastics also can leach harmful chemicals into the water which can be carried along the food chain back to

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: light refraightment refightment refightment refightment refightment refightment refightment refightment refightment refightment refightment refightment refightment refightment refightment ref

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

    based on the borrower's income, expenses, and credit history. The results of a stress test can impact the amount a borrower is able to borrow for a mortgage and the interest rate they receive.⏎Debt Servicing Ratio

  2. #2 peak

    how much interest to charge. By not knowing their credit score, people risk paying more interest when they take out a loan and may face limits on the amount they can borrow. In extreme circumstances, they may find themselves “locked out” of credit

  3. #3 peak

    also known as the “Poop to Power” project. The station will provide fuel for the first natural gas vehicles in the City’s fleet. Initially, vehicles will fill up on natural gas from conventional sources, but once the full project is complete

  4. #4 peak

    nett Town site Company dedicated the streets, avenues and alleys to the use of the public. The right and privilege to construct gas, water and sewer lines, to erect telephone and electric light poles was reserved to the company. A small town site

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. L23:53 excites
  2. L23:373 suppresses 'to' after a noun or adjective
  3. L22:627 suppresses
  4. L19:2632 suppresses
  5. L23:491 excites
  6. L17:645 excites
  7. L11:223 suppresses
  8. L17:1361 suppresses 'that' after verbs like recommend
  9. L4:113 suppresses
  10. L6:1133 excites "to" before an infinitive verb
  11. L21:536 excites
  12. L16:1153 weakly suppresses
  13. L7:520 weakly excites
  14. L18:178 weakly suppresses
  15. L20:349 weakly excites
  16. L19:1830 weakly excites
  17. L18:971 weakly suppresses
  18. L9:226 weakly suppresses 'to' starting an infinitive
  19. L15:184 weakly suppresses
  20. L7:681 weakly excites 'from' after prevent or stop
  21. L21:1663 weakly excites
  22. L21:1657 weakly suppresses
  23. L0:1018 weakly excites
  24. L6:1880 weakly excites "by" introducing a deadline or means

Feeds later layers

  1. L27:2060 weakly suppresses
  2. L27:3028 weakly suppresses
  3. L27:1972 weakly excites
  4. L27:1328 weakly suppresses
  5. L27:2991 weakly excites
  6. L27:2653 weakly suppresses
  7. L27:1437 weakly suppresses
  8. L27:1414 weakly excites
  9. L27:2082 weakly excites
  10. L27:1048 weakly suppresses
  11. L27:1705 weakly excites
  12. L27:973 weakly suppresses
  13. L27:1041 weakly excites
  14. L27:3008 weakly excites
  15. L27:1231 weakly suppresses
  16. L27:1471 weakly suppresses
  17. L27:330 weakly excites
  18. L27:2322 weakly excites
  19. L27:560 weakly excites
  20. L27:1341 weakly excites
  21. L27:2519 weakly excites
  22. L27:1762 weakly excites
  23. L27:2816 weakly excites
  24. L27:2111 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.

1 values between -43.91 and -42.051 values between -34.61 and -32.752 values between -25.32 and -23.462 values between -23.46 and -21.601 values between -21.60 and -19.744 values between -19.74 and -17.886 values between -17.88 and -16.026 values between -16.02 and -14.1611 values between -14.16 and -12.3018 values between -12.30 and -10.4519 values between -10.45 and -8.5938 values between -8.59 and -6.7354 values between -6.73 and -4.87112 values between -4.87 and -3.01297 values between -3.01 and -1.152441 values between -1.15 and 0.71501 values between 0.71 and 2.57155 values between 2.57 and 4.4398 values between 4.43 and 6.2943 values between 6.29 and 8.1537 values between 8.15 and 10.0016 values between 10.00 and 11.8613 values between 11.86 and 13.728 values between 13.72 and 15.589 values between 15.58 and 17.445 values between 17.44 and 19.304 values between 19.30 and 21.161 values between 21.16 and 23.021 values between 24.88 and 26.742 values between 26.74 and 28.592 values between 28.59 and 30.45offnegativepositiveoften

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. 地上
  2. 思う
  3. perish
  4. |_|
  5. xfc
  6. 地说
  7. xFFFF
  8. 途
  9. Err
  10. 床上

Suppresses

  1. чь
  2. .depth
  3. ight
  4. edReader
  5. ző
  6. 蜇
  7. .char
  8. bbing
  9. bay
  10. rite