Neuron Explorer

Qwen3-0.6B

'F' starting a word in a split

A neuron in layer 27 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 a capital 'F' that begins a word, especially in lists of translations or after a line break, such as 'Frecuencia' or 'FPGA'. Turning it up made the model repeat themes like 'picture' or 'growing', and turning it down made it repeat 'air' or 'chart'.

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

    . Fixtures are designed to hold the workpiece in a specific position and orientation, allowing for consistent and accurate processing.⏎Deutsch: Frequenz / Español: Frecuencia / Português: Frequéncia / Français: Frêquence

  2. #2 peak

    . The period is the duration of one cycle in a repeating event, so the period is the reciprocal of the frequency.⏎Deutsch: Fabrik / Español: Fábrica / Português: Fábrica / Français: Us

  3. #3 peak

    defensores for the frontier. In the principal northern towns four squadrons and one company were to be formed under the name of "Defensores de la Frontera." Each squadron was to be under the command of a "commandant

  4. #4 peak

    when the bow tends to rise. tunnel, water-See WATER TUNNEL. tunnel, wind-See WIND TUNNEL. turbulent flow-See F

  5. #5 peak

    controller, an FPGA is not designed to execute a specific set of instructions. Instead, developers program the FPGA to perform specific logic functions by configuring its hardware components. FPGAs consist of a matrix of configurable logic

  6. #6 peak

    enable developers to design digital circuits and systems at a higher level of abstraction, making FPGA design more efficient and less error-prone.⏎Differences Between Microcontrollers and FPGAs⏎Microcontrollers are programmable, which means that developers can write software

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: after recharting chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart chart

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

    these mysterious disappearances by locating their last names in the word search puzzle in our Missing Persons worksheet. If you uncover all eight names, the FBI (Floss Bureau of Investigation) will reward you with the title of Master

  2. #2 peak

    where they seem not to have learned one of the great lessons of 20th century art, illustrated by Marcel Duchamp’s 1917 “Fountain”: that the intent of the artist is what makes something art.⏎Origin of

  3. #3 peak

    ptomatically.⏎-Please note this is not a medical mask and should not be worn as personal protective equipment.⏎-The face covering does not guarantee safety or protection.⏎-Face coverings should not be used as a replacement for social distancing or self isolation

  4. #4 peak

    1,800 illustrated dictionary⏎entries. Most entries have links to additional information, activities,⏎and child-friendly, related websites. There are links to English-French,⏎English-German, English-Portuguese, or the English-S

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. L23:2441 excites
  2. L22:369 excites
  3. L18:437 excites
  4. L20:211 excites
  5. L18:328 suppresses
  6. L2:1622 weakly excites
  7. L21:2605 weakly excites
  8. L24:2737 weakly excites
  9. L3:1098 weakly excites
  10. L1:9 weakly excites
  11. L21:237 weakly suppresses
  12. L21:381 weakly suppresses zeros inside long number codes
  13. L21:2432 weakly excites
  14. L4:1892 weakly excites
  15. L6:13 weakly excites
  16. L21:2484 weakly suppresses
  17. L26:1564 weakly excites
  18. L17:378 weakly excites
  19. L13:1539 weakly excites
  20. L3:412 weakly excites
  21. L23:2013 weakly suppresses
  22. L1:2099 weakly excites
  23. L22:151 weakly suppresses
  24. L4:2660 weakly excites

Feeds later layers

None: this is the last layer.

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 -27.03 and -25.411 values between -20.54 and -18.922 values between -17.30 and -15.682 values between -15.68 and -14.061 values between -14.06 and -12.431 values between -10.81 and -9.191 values between -9.19 and -7.574 values between -7.57 and -5.957 values between -5.95 and -4.3317 values between -4.33 and -2.7029 values between -2.70 and -1.083564 values between -1.08 and 0.54257 values between 0.54 and 2.1612 values between 2.16 and 3.782 values between 3.78 and 5.411 values between 5.41 and 7.031 values between 7.03 and 8.651 values between 8.65 and 10.271 values between 11.89 and 13.521 values between 15.14 and 16.761 values between 18.38 and 20.001 values between 36.22 and 37.84offnegativepositiveoften

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. icut
  2. {}",
  3. -paper
  4. successor
  5. ania
  6. low
  7. 荼
  8. createCommand
  9. opleft
  10. lowest

Suppresses

  1. essian
  2. nhuận
  3. 弹性
  4. Clayton
  5. arding
  6. ventions
  7. throw
  8. repent
  9. ChildIndex
  10. nit