Neuron Explorer

Qwen3-0.6B

end-of-document token after a web snippet

A neuron in layer 11 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

Fires weakly on the invisible marker that the model uses to show one web page or document has ended and another begins. Turning it up made the model produce hesitant, rambling "I mean" text.

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

    dark brown (2).⏎- Length: 11 cm (2)⏎- 6 g (2)<|endoftext|>Causes of Cancer⏎Cancers can develop from a variety of sources and causes

  2. #2 peak

    per minute|⏎|45 years||88-149 beats per minute||175 beats per minute|⏎|50 years||85<|endoftext|>A Brief Introduction

  3. #3 peak

    : ||9781312744967 (DRM-EPUB)|⏎|Read Aloud: ||not allowed|<|endoftext|>In 1979, police in Pensacola, Florida, arrested

  4. #4 peak

    .org/catalog/V_4E4EE3EAD24043E2BC0B972CD24B176C<|endoftext|>Rurik DynastyArticle Free Pass⏎Rurik Dynasty, princes of K

  5. #5 peak

    2021⏎- Nepali Climbers Open New Trekking Route on Mt. Kilimanjaro - September 9, 2021<|endoftext|>In this VB.NET sample project, you will learn to create a simple dictionary.

  6. #6 peak

    In The Moon"⏎- 'Star Trek: The Next Generation' - "Rascals"⏎- 'Warehouse 13' - "Time Will Tell"<|endoftext|>I question why "pres" was marked incorrect in this exercise. As I understand

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: **************************************************

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

    J.M.; Mortimer, S.R.; Brereton, T.; Redhead, J.W.; Thomas, J.A.; Pywell, R.F.. 20

  2. #2 peak

    different. Although the latest legs have exceeded all expectations in stress-testing–the 3-D-printed plastic is about as strong as a traditional prosthetic-–coons and associate professor Matt Ratto have yet to find the perfect thickness, the

  3. #3 peak

    2.16.128.0/17 SUBNET3 : 10101100.00010001.0yyyyyyy.yyyyyyyy => 172.1

  4. #4 peak

    /Share-Alike License⏎- n. An underwater explorer.⏎from WordNet 3.0 Copyright 2006 by Princeton University. All rights reserved.⏎- n. an underwater swimmer equipped with a face mask and foot fins

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. L9:15 strongly excites
  2. L5:25 strongly excites
  3. L10:28 strongly suppresses unclear: weak, articles before nouns
  4. L5:36 suppresses
  5. L10:51 excites
  6. L5:112 excites
  7. L5:498 suppresses
  8. L10:150 suppresses
  9. L7:53 weakly excites
  10. L2:69 weakly excites
  11. L0:18 weakly suppresses
  12. L6:7 weakly suppresses
  13. L2:75 weakly suppresses
  14. L6:13 weakly excites
  15. L8:10 weakly excites
  16. L3:12 weakly excites 'and' joining the last item in a list
  17. L3:2870 weakly excites
  18. L5:110 weakly suppresses
  19. L5:46 weakly excites
  20. L6:24 weakly excites
  21. L3:115 weakly suppresses
  22. L8:14 weakly excites
  23. L6:18 weakly excites
  24. L10:895 weakly suppresses

Feeds later layers

  1. L13:654 strongly suppresses
  2. L13:105 strongly excites
  3. L14:847 strongly suppresses
  4. L14:68 strongly excites
  5. L14:210 strongly suppresses
  6. L15:892 strongly suppresses
  7. L13:2270 strongly excites
  8. L15:62 strongly excites
  9. L12:1574 strongly excites
  10. L12:396 strongly excites
  11. L15:514 strongly excites
  12. L15:907 strongly suppresses the '0' in numbers like 30 or 40
  13. L16:368 strongly suppresses
  14. L16:192 strongly suppresses
  15. L12:2159 strongly suppresses
  16. L12:1307 suppresses
  17. L18:942 suppresses
  18. L16:841 excites
  19. L12:741 suppresses
  20. L14:98 excites
  21. L14:2781 excites
  22. L15:1550 suppresses
  23. L15:13 suppresses
  24. L22:213 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 -0.74 and -0.711 values between -0.65 and -0.623 values between -0.59 and -0.561 values between -0.56 and -0.538 values between -0.53 and -0.5010 values between -0.50 and -0.4724 values between -0.47 and -0.4437 values between -0.44 and -0.4146 values between -0.41 and -0.3756 values between -0.37 and -0.3484 values between -0.34 and -0.31133 values between -0.31 and -0.28157 values between -0.28 and -0.25210 values between -0.25 and -0.22276 values between -0.22 and -0.19310 values between -0.19 and -0.16333 values between -0.16 and -0.13350 values between -0.13 and -0.10366 values between -0.10 and -0.07329 values between -0.07 and -0.04272 values between -0.04 and -0.01333 values between -0.01 and 0.02168 values between 0.02 and 0.05142 values between 0.05 and 0.0893 values between 0.08 and 0.1160 values between 0.11 and 0.1538 values between 0.15 and 0.1823 values between 0.18 and 0.2114 values between 0.21 and 0.249 values between 0.24 and 0.274 values between 0.27 and 0.304 values between 0.30 and 0.336 values between 0.33 and 0.362 values between 0.36 and 0.392 values between 0.39 and 0.422 values between 0.42 and 0.451 values between 0.45 and 0.48offnegativepositiveoften

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. отор
  2. ford
  3. ods
  4. dr
  5. dr
  6. utex
  7. Constructed
  8. [@
  9. utan
  10. onna

Suppresses

  1. grou
  2. Pinterest
  3. bä
  4. ält
  5. 路线
  6. ACHER
  7. Pin
  8. â
  9. →person
  10. péri