Neuron Explorer

Qwen3-0.6B

gap where a number was removed

A neuron in layer 9 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 empty spots in text where a number or symbol seems to have been stripped out, such as "Volume 54, Issue [blank] 5". Turning it up did not make the writing clearly different.

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

    bright, mocking beauty.⏎October 2003 | Volume 54, Issue 5⏎Your example of occupy brings to mind the way in which new words are

  2. #2 peak

    link="none" columns="2" size="medium" ids="2568,2567,25

  3. #3 peak

    Nursery and Reception) and the National Curriculum for KS1 (in Years 1 and 2).⏎Music National Curriculum for KS1⏎Key stage 1

  4. #4 peak

    other.’”⏎Most Circassians (sur-CASH-ins) took refuge in the lands of the Ottoman Empire. Today, 2 million of the world’s 7 million Circassians live in Turkey, with another 120,

  5. #5 peak

    Search for questions⏎Still looking for answers? Try searching for what you seek or ask your own question.⏎Posted 31 Aug 2009 • 1 answer⏎Posted 20 Jan 2010 • 1

  6. #6 peak

    support students with dyslexia or other reading challenges, including screening students from kindergarten through high school and creating targeted programs at 160 of the city’s 1,600 schools.⏎Read the

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: I have to be honest with you, I have to be honest with you, I have to be honest with you, I have to be honest with you, I have to be honest with you, I have to be honest with you,

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

    Tue, 30 Oct 2012, 11:07:11 EDT⏎- How Martian winds make rocks walkThu, 8 Jan 2009, 14:01:0

  2. #2 peak

    35 - HMS Victory And HMS Dreadnought, 2 x £1.55 - HMS Warrior and Sovereign Of The Seas, 2 x £1.60 - HMS King George V and HMS Beagle).⏎There’s

  3. #3 peak

    download Quicktime; 1.6 Mb)⏎B. Convection in action (download .mp4; 22 Mb)⏎C. Clean-up (download .mp4; 350 Kb)⏎Model 2: Candle,

  4. #4 peak

    2008 US$ 12.00 176 pages⏎Teacher Evaluation to Enhance Professional Practice 2000 US$ 24.95 167 pages<|endoftext|>Gate is the national monument

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. L7:8 strongly excites
  2. L7:207 excites
  3. L8:27 suppresses
  4. L8:282 suppresses
  5. L8:177 excites
  6. L7:7 excites
  7. L8:21 suppresses
  8. L8:111 suppresses the space before a year in dates
  9. L6:228 excites comma after a filler word
  10. L6:443 suppresses
  11. L7:135 excites 'of' after 'issue' in citations
  12. L6:296 suppresses
  13. L6:325 excites
  14. L6:429 suppresses
  15. L6:1741 excites
  16. L8:79 excites
  17. L3:180 suppresses
  18. L6:1105 excites 'that' after a finding or surprise verb
  19. L6:1031 suppresses
  20. L3:49 excites
  21. L6:436 excites
  22. L6:423 suppresses
  23. L6:159 excites
  24. L6:402 suppresses

Feeds later layers

  1. L14:830 suppresses
  2. L10:40 excites
  3. L17:1096 suppresses
  4. L15:202 suppresses mixed, digits in URLs and comparisons
  5. L10:459 suppresses
  6. L13:25 excites
  7. L13:909 suppresses 'to' before a verb after a noun
  8. L16:297 suppresses
  9. L13:915 suppresses
  10. L12:652 suppresses
  11. L13:376 suppresses
  12. L14:672 suppresses
  13. L14:307 suppresses hyphen inside phone numbers
  14. L16:269 excites
  15. L17:327 suppresses
  16. L18:156 suppresses
  17. L16:223 excites
  18. L15:1381 suppresses
  19. L18:606 suppresses
  20. L17:12 excites
  21. L19:93 excites
  22. L13:21 suppresses
  23. L19:1129 suppresses
  24. L16:1380 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 -1.96 and -1.872 values between -1.79 and -1.701 values between -1.53 and -1.451 values between -1.36 and -1.281 values between -1.28 and -1.191 values between -1.11 and -1.021 values between -0.94 and -0.851 values between -0.85 and -0.773 values between -0.77 and -0.681 values between -0.60 and -0.511 values between -0.51 and -0.4310 values between -0.43 and -0.3464 values between -0.34 and -0.26218 values between -0.26 and -0.17594 values between -0.17 and -0.091242 values between -0.09 and -0.001046 values between -0.00 and 0.08384 values between 0.08 and 0.17186 values between 0.17 and 0.2576 values between 0.25 and 0.3433 values between 0.34 and 0.4221 values between 0.42 and 0.518 values between 0.51 and 0.592 values between 0.59 and 0.683 values between 0.68 and 0.763 values between 0.76 and 0.852 values between 0.85 and 0.931 values between 1.10 and 1.191 values between 1.36 and 1.44offnegativepositiveoften

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. Ba
  2. Dise
  3. Reach
  4. 福利
  5. 裸
  6. ника
  7. cern
  8. 作者所有
  9. 情怀
  10. gp

Suppresses

  1. ước
  2. HostException
  3. entieth
  4. orestation
  5. 看一看
  6. understood
  7. ソン
  8. 理解和
  9. edelta
  10. ürn