Neuron Explorer

Qwen3-0.6B

unclear: weak mid-phrase prepositions

A neuron in layer 16 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

Hard to pin down. Its weak highs fall on small connecting words such as "the", "on" and "of" in the middle of a sentence, and its strongest negative hits are on words like "manage" and "to" in comparisons. Steering did not show a clear theme.

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

    ) "The gun fired"⏎<|endoftext|>NASA is trying to decide whether to continue preparing the space shuttle Atlantis for a launch attempt on Tuesday or to roll it back into the Vehicle Assembly Building (VAB) to take

  2. #2 peak

    ate Acids and Bases practice, you can also practice Conjugate Acids and Bases practice problems.⏎What professor is this problem relevant for?⏎Based on our data, we think this problem is relevant for Professor Finnegan's class at

  3. #3 peak

    or Center for Neuromodulation⏎What is Parkinson's disease?⏎Parkinson's disease (PD or, simply, Parkinson's) is the most common form of parkinsonism, a group of motor system disorders.⏎It is a slowly progressing

  4. #4 peak

    ynursingpaper.com/wp-content/uploads/2020/06/cropped-155972031988413967-1-768x146-1

  5. #5 peak

    . Lythronax stands out from its contemporaries in having a much wider skull at the eyes and a narrow short snout, similar to its relative T. rex, which lived 10-12 million years later. Dr.

  6. #6 peak

    homes, from the ground, through sump pumps, cracks in the foundation, etc., it is considered a health hazard.⏎Any of the beams that slope from the ridge of a roof to the eaves

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: the other... ,... the other... . ,... the... ... the... ... the... ... the... ... the... ... the... ... 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

    Trim The Roots Of A Bonsai Tree? (Step by Step)⏎Caring for bonsai trees is a unique art form, different from how you would manage other plants.⏎Looking after non-bonsai plants involves re-potting them

  2. #2 peak

    no one better. He wrote more letters to her than he wrote to anyone. All her life, she wrote back: letter after letter filled with news

  3. #3 peak

    ⏎The debate over whether to use dashes or underscores to represent spaces in URLs is rather heated in the web development community, but not quite as extremely so as that of whether to use tabs or spaces when indenting code. So, what is better

  4. #4 peak

    is clicked.<|endoftext|>Americans with Disabilities Act (ADA)⏎The Americans with Disabilities Act (ADA) provides civil rights protections to individuals with disabilities similar to those provided to individuals

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. L15:2981 strongly suppresses
  2. L15:2703 suppresses
  3. L14:2076 suppresses
  4. L13:2320 excites
  5. L13:1692 excites
  6. L14:1991 excites
  7. L15:2067 excites
  8. L13:1991 excites
  9. L14:733 suppresses
  10. L14:1741 excites
  11. L14:2727 excites
  12. L13:2483 suppresses
  13. L15:1738 excites
  14. L14:399 excites
  15. L14:2595 suppresses
  16. L15:2223 excites
  17. L15:2248 suppresses
  18. L15:2503 excites
  19. L14:1042 excites
  20. L15:476 excites
  21. L13:2312 excites
  22. L14:1537 excites
  23. L15:1950 excites
  24. L14:978 suppresses

Feeds later layers

  1. L17:239 suppresses
  2. L17:807 excites
  3. L18:825 suppresses 'as well as' and list connectors
  4. L18:1536 suppresses end of web page disclaimer
  5. L18:1850 suppresses
  6. L18:556 excites
  7. L17:1586 suppresses
  8. L18:206 excites
  9. L19:1567 suppresses
  10. L19:355 excites
  11. L17:174 excites
  12. L19:1673 suppresses
  13. L20:697 weakly suppresses
  14. L17:1109 weakly suppresses
  15. L22:73 weakly excites
  16. L20:204 weakly suppresses
  17. L20:1189 weakly suppresses
  18. L19:401 weakly excites
  19. L24:174 weakly suppresses
  20. L20:708 weakly excites
  21. L18:540 weakly suppresses
  22. L20:2848 weakly suppresses
  23. L25:935 weakly suppresses
  24. L17:1437 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 -4.69 and -4.531 values between -2.96 and -2.801 values between -2.64 and -2.482 values between -2.33 and -2.172 values between -2.17 and -2.012 values between -2.01 and -1.852 values between -1.85 and -1.702 values between -1.54 and -1.388 values between -1.38 and -1.236 values between -1.23 and -1.0715 values between -1.07 and -0.9130 values between -0.91 and -0.7547 values between -0.75 and -0.6086 values between -0.60 and -0.44178 values between -0.44 and -0.28394 values between -0.28 and -0.122003 values between -0.12 and 0.03798 values between 0.03 and 0.19186 values between 0.19 and 0.3574 values between 0.35 and 0.5133 values between 0.51 and 0.6616 values between 0.66 and 0.8211 values between 0.82 and 0.983 values between 0.98 and 1.145 values between 1.14 and 1.291 values between 1.29 and 1.451 values between 1.45 and 1.61offnegativepositiveoften

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. rosse
  2. -public
  3. Ƣ
  4. ieurs
  5. oğlu
  6. imized
  7. }> ⏎
  8. 轸
  9. łow
  10. dney

Suppresses

  1. other
  2. 其他
  3. (other
  4. 其他的
  5. other
  6. main
  7. footnote
  8. 得多
  9. 其它
  10. Other