Neuron Explorer

Qwen3-0.6B

'less' in 'colorless'

A neuron in layer 2 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

This neuron mostly lights up on the ending 'less' in words like 'colorless', and on a few other word pieces. Turning it up made the model's writing drift toward unrelated topics, so the link is only loose.

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

    of incidence at large angles and a flatter response to different wavelenghts over the visible. Consequently, graded-index films are desirable in automotive applications where colorless coatings are required. A sol-gel process is presented for the formation of a

  2. #2 peak

    can eventually lead to structural damage in your home if left untreated. A dehumidifier or air conditioner can help.⏎Answer: True. Radon—a colorless, tasteless and odorless gas—seeps into homes through cracks in walls

  3. #3 peak

    often looks “cloudy”, tabular rectangle shape||Crystals are often half grey and half black.|⏎|Plagioclase feldspar||Colourless, sometimes looks cloudy, tabular rectangle shape||Stripy like a zebra

  4. #4 peak

    ements are manufactured to meet the specifications of ASTM C150. ASTM C150 provides for ten types of portland cement: Type INormal Type IANormal, air-entraining Type IIModerate sulfate resistance Type II

  5. #5 peak

    gases in the Periodic Table. It is the heaviest known gas, is radioactive, and is essentially inert. At ordinary temperatures, radon is a colorless gas, but when it is cooled below its freezing point it gains a stunning phosph

  6. #6 peak

    eggs. They are colorless and about 0.1 mm in diameter. The eggs are negatively buoyant

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: 2023 - present I am a woman: I am from South Korea, and my parents are Chinese? I am a woman: I am from South Korea, and my parents are Chinese? I am a woman

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

    hold their breath underwater for extended periods, making them excellent divers. The duration of a turtle’s breath-holding capabilities varies depending on several factors, including the species of turtle, water temperature, and activity level. While some species of turtles can hold

  2. #2 peak

    Because of this, an arson investigator’s most important witnesses are the firefighters that first arrive on the scene. The investigator will ask firefighters for details such as the color of the smoke, damaged sprinkler systems, condition of doors and windows, and general

  3. #3 peak

    is a class of diseases characterized by out-of-control cell growth. There are over 100 different types of cancer, and each is classified by the type of cell that is initially affected.⏎Cancer harms the body when damaged cells divide uncont

  4. #4 peak

    under federal law.⏎Legal Liability for School-bus Accidents⏎Legal liability for injuries sustained by your child in a school-bus accident may depend on the type of parties involved.⏎

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. L1:2801 weakly excites
  2. L0:2038 weakly excites
  3. L0:1240 weakly suppresses
  4. L0:1515 weakly suppresses
  5. L0:1631 weakly excites
  6. L1:653 weakly suppresses
  7. L1:626 weakly suppresses
  8. L0:174 weakly suppresses
  9. L1:751 weakly suppresses
  10. L0:535 weakly excites
  11. L1:1949 weakly excites
  12. L1:2342 weakly excites
  13. L1:464 weakly excites
  14. L1:166 weakly suppresses
  15. L0:1507 weakly suppresses
  16. L1:74 weakly excites
  17. L0:158 weakly excites
  18. L0:2194 weakly suppresses
  19. L0:642 weakly excites
  20. L1:2626 weakly excites
  21. L0:2579 weakly suppresses
  22. L0:2374 weakly excites
  23. L1:1435 weakly excites 'as' in 'just as'
  24. L1:2942 weakly excites

Feeds later layers

  1. L5:829 suppresses the word 'age' and 'how many'
  2. L7:731 suppresses
  3. L4:340 suppresses
  4. L9:153 suppresses
  5. L13:2618 suppresses
  6. L7:1186 suppresses
  7. L7:2414 suppresses
  8. L6:240 suppresses
  9. L13:811 weakly excites
  10. L12:660 weakly suppresses
  11. L6:447 weakly suppresses
  12. L19:1460 weakly suppresses
  13. L9:942 weakly suppresses
  14. L4:2826 weakly suppresses
  15. L3:382 weakly excites
  16. L4:297 weakly suppresses
  17. L15:1566 weakly suppresses
  18. L5:430 weakly suppresses
  19. L8:882 weakly suppresses
  20. L9:155 weakly excites
  21. L5:121 weakly suppresses
  22. L11:425 weakly suppresses
  23. L4:638 weakly suppresses
  24. L11:778 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 -2.93 and -2.841 values between -2.38 and -2.291 values between -1.84 and -1.751 values between -1.75 and -1.653 values between -1.47 and -1.383 values between -1.11 and -1.021 values between -1.02 and -0.923 values between -0.92 and -0.835 values between -0.83 and -0.742 values between -0.74 and -0.656 values between -0.65 and -0.564 values between -0.56 and -0.4717 values between -0.47 and -0.3826 values between -0.38 and -0.2886 values between -0.28 and -0.19267 values between -0.19 and -0.101070 values between -0.10 and -0.011651 values between -0.01 and 0.08551 values between 0.08 and 0.17163 values between 0.17 and 0.2633 values between 0.26 and 0.367 values between 0.36 and 0.451 values between 0.45 and 0.543 values between 0.54 and 0.632 values between 0.63 and 0.72offnegativepositiveoften

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. compound
  2. _COMPLETED
  3. going
  4. Compound
  5. 的信心
  6. compound
  7. ’à
  8. görd
  9. lad
  10. 成功的

Suppresses

  1. territ
  2. traits
  3. 擤
  4. emap
  5. Territories
  6. tracked
  7. 旬
  8. CharacterSet
  9. /title
  10. etails