Neuron Explorer

Qwen3-0.6B

blank spot before a year

A neuron in layer 16 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 is active on the empty space just before a year or a number in copyright lines and dates, such as 'Copyright © 1997' or 'died January 29, 1963'. The activations are fairly weak. Turning it up or down did not produce a clear change in the writing.

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

    more propositions that cannot be true simultaneously.⏎b. two or more attributes that cannot simultaneously belong to the same object.⏎Random House Unabridged Dictionary, Copyright © 1997, by Random House, Inc., on Infoplease.

  2. #2 peak

    - Courier Delivery: Yes⏎- Store Delivery: Yes⏎Prices are for internet purchases only. Prices and availability in WHSmith Stores may vary significantly⏎© Copyright 2013 - 2017 WHSmith and its suppliers.⏎

  3. #3 peak

    Robert Lee Frost, (born March 26, 1874, San Francisco, California, U.S.—died January 29, 1963, Boston, Massachusetts), American poet who was much admired for

  4. #4 peak

    advice of his mother especially in Proverbs 31: 3 Do not spend your strength on women, your vigor on those who ruin kings

  5. #5 peak

    MedicineHealth Medical Reference from Healthwise⏎To learn more visit Healthwise.org⏎© 1995-2012 Healthwise, Incorporated. Healthwise

  6. #6 peak

    mus and Strabismus. I know you guys just love scientific names!⏎<|endoftext|>- George of Trebizond⏎- (1396-ca. 1472; Latin name, Trapezuntius)Humanist and

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 to to to to 1 to 1 3 to 1 1 1 1, 1 1 1, 1 1 1

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

    NGAGE. That means walk away – “We'll pick this up later, when we have a chance to cool down – it’ll be better.” You always have that choice.⏎On another level, if YOU are

  2. #2 peak

    ised the 16 sixteenth notes as they go by and it is much easier than saying ‘ 1 2 3 4 5 6 7 8 9 10 11 12 1

  3. #3 peak

    172.16.0.0/12 => min size for network id = 12 max size for host id = 20 10.0.0.0/8 => max size for host id

  4. #4 peak

    have found disturbing concentrations of pesticides in some baby foods. And the U.S. Department of Agriculture has found weed killers widespread in finished tap water. Environmentalists have had important successes in forcing pesticides that presented unacceptably high dietary risks off the market

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. L2:239 strongly suppresses
  2. L4:2532 strongly excites surname pieces split across tokens
  3. L0:258 strongly excites
  4. L2:1417 strongly suppresses
  5. L1:1256 strongly excites
  6. L2:964 strongly suppresses
  7. L1:831 strongly suppresses
  8. L4:185 strongly excites
  9. L1:217 strongly suppresses
  10. L6:1699 suppresses
  11. L6:394 excites
  12. L4:53 excites blank slot before a number
  13. L0:465 suppresses
  14. L5:1896 suppresses
  15. L1:95 excites
  16. L0:667 excites
  17. L2:404 excites
  18. L2:1607 suppresses
  19. L7:432 suppresses
  20. L6:715 excites
  21. L0:1916 excites
  22. L5:139 excites
  23. L4:376 suppresses
  24. L6:2064 excites

Feeds later layers

  1. L18:649 strongly suppresses
  2. L18:122 suppresses
  3. L25:422 suppresses
  4. L20:396 suppresses the D in 3D or 2D
  5. L19:581 suppresses blank slot before a number
  6. L18:185 excites
  7. L26:2003 suppresses hyphen joining two word parts
  8. L22:2384 suppresses
  9. L17:148 excites
  10. L22:824 suppresses blank before an expected number
  11. L22:372 suppresses empty spot before a number or "3D"
  12. L20:2800 excites
  13. L21:339 suppresses spot just before a year
  14. L18:1995 excites
  15. L20:344 suppresses hyphen inside compound words
  16. L26:1566 excites
  17. L27:2439 suppresses slash inside web addresses
  18. L27:170 suppresses unclear: spaces in mixed-language text
  19. L26:254 weakly excites
  20. L17:285 weakly suppresses
  21. L19:1304 weakly excites
  22. L27:327 weakly excites
  23. L19:265 weakly excites
  24. L24:248 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.

2 values between -3.93 and -3.751 values between -3.38 and -3.202 values between -3.20 and -3.012 values between -3.01 and -2.834 values between -2.83 and -2.654 values between -2.65 and -2.464 values between -2.46 and -2.2811 values between -2.28 and -2.1011 values between -2.10 and -1.9126 values between -1.91 and -1.7328 values between -1.73 and -1.5530 values between -1.55 and -1.3755 values between -1.37 and -1.1891 values between -1.18 and -1.00106 values between -1.00 and -0.82146 values between -0.82 and -0.63204 values between -0.63 and -0.45234 values between -0.45 and -0.27266 values between -0.27 and -0.08418 values between -0.08 and 0.10380 values between 0.10 and 0.28422 values between 0.28 and 0.47510 values between 0.47 and 0.65530 values between 0.65 and 0.83322 values between 0.83 and 1.0284 values between 1.02 and 1.2011 values between 1.20 and 1.381 values between 1.38 and 1.562 values between 1.56 and 1.751 values between 3.21 and 3.40offnegativepositiveoften

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. iance
  2. hom
  3. /output
  4. kind
  5. -kind
  6. i
  7. item
  8. ⏎
  9. ⏎⏎
  10. living

Suppresses

  1. erset
  2. phans
  3. [op
  4. aldi
  5. 是他
  6. .FILES
  7. (Size
  8. nog
  9. 蕻
  10. 团长