Neuron Explorer

Qwen3-0.6B

document-end markers, mixed

A neuron in layer 12 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 mostly fires at the hidden marker where one web page ends and another begins, and sometimes on a word at the end of a sentence. Its negative side reacts to odd spots like 'NC' in a licence tag. Turning it up gave repetitive text with no 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

    for an accident may be limited by governmental immunity, even if the accident was caused by the negligence of its employee, the bus driver.⏎When governmental immunity applies,<|endoftext|>Actinopteri (ray-finned fishes) > Cichliformes

  2. #2 peak

    .⏎For severe low blood sugar, your child may need a glucagon injection. Ask your child’s healthcare provider about glucagon emergency kits for home and school.<|endoftext|>|This article needs additional citations for verification. (July 2012

  3. #3 peak

    least 12 missions to carry cargo to and from the International Space Station as part of the Commercial Resupply Services (CRS) contract for NASA.⏎For<|endoftext|>|1st person||me me||nos us|⏎|2nd person||

  4. #4 peak

    applications include improved communications technology, television displays and night-vision devices.⏎Tags: Lene Vestergaard Hau, physicist, spped of light.<|endoftext|>The Real

  5. #5 peak

    to do better in tests when they used these learning materials”.⏎While the data generated as a byproduct of online courses provide a valuable resource in their own right<|endoftext|>The Causal Argument⏎Choose your event

  6. #6 peak

    . You can communicate a thought, but your thought may not be understood. In some cases, your thought may not even reach the proper target. Thats why it pays to ask questions to make certain

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: (a) a certain piece of (a) is a certain piece of (a) that is, (a) is... a certain piece of... a (a) as, the... (a) is... (a) is

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

    as a facebook live on 15th october download a worksheet here download the october to november schedule in colour download the october to november schedule in.⏎Vocabulary Worksheet Mix Set context pre teach vocabulary these stages go online for

  2. #2 peak

    Fung Tjin, Erqian Wang and Yifan Zhang and directed by A. J. Hildebrand.)⏎Open content licensed under CC BY-NC-SA⏎Given an integer , the exponential sum version of the Gauss sum modulo is

  3. #3 peak

    in Washington, D. C. (Image via Zach Montellaro on Flickr, CC BY-NC-SA 2.0)⏎The two Supreme Court cases, Jones v. City

  4. #4 peak

    equated as being the American response to the national epic.⏎2 Books referred to as "Great American Novel",⏎3 See also,⏎5 Further reading,⏎6 External links,⏎While fiction was written in colonial America as early as the 17th

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. L8:3 suppresses
  2. L11:2 excites unclear: blank before a year or number
  3. L9:5 suppresses
  4. L8:13 excites
  5. L8:7 suppresses
  6. L5:86 excites
  7. L8:4 suppresses
  8. L10:65 excites
  9. L8:10 suppresses
  10. L7:122 suppresses
  11. L8:16 excites
  12. L5:60 suppresses
  13. L8:12 suppresses
  14. L10:0 suppresses
  15. L8:1 weakly excites
  16. L8:293 weakly suppresses
  17. L11:128 weakly excites
  18. L5:216 weakly excites
  19. L7:62 weakly suppresses
  20. L10:507 weakly suppresses
  21. L3:464 weakly suppresses
  22. L6:12 weakly suppresses
  23. L6:7 weakly excites
  24. L9:223 weakly suppresses

Feeds later layers

  1. L14:2071 weakly suppresses
  2. L14:2600 weakly suppresses
  3. L22:775 weakly suppresses
  4. L17:489 weakly excites
  5. L15:2250 weakly suppresses
  6. L14:2974 weakly suppresses
  7. L13:456 weakly excites
  8. L14:110 weakly excites
  9. L21:391 weakly suppresses
  10. L16:1695 weakly suppresses
  11. L14:416 weakly excites
  12. L23:1640 weakly suppresses
  13. L27:243 weakly excites
  14. L16:2393 weakly suppresses
  15. L19:2471 weakly suppresses
  16. L18:1366 weakly excites
  17. L15:2615 weakly suppresses
  18. L15:73 weakly excites
  19. L25:500 weakly excites
  20. L18:2980 weakly excites
  21. L16:1961 weakly excites
  22. L13:2963 weakly suppresses
  23. L26:586 weakly suppresses
  24. L18:902 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 -3.58 and -3.431 values between -2.65 and -2.502 values between -2.50 and -2.355 values between -2.35 and -2.193 values between -2.19 and -2.044 values between -2.04 and -1.883 values between -1.88 and -1.738 values between -1.73 and -1.5718 values between -1.57 and -1.4233 values between -1.42 and -1.2666 values between -1.26 and -1.1187 values between -1.11 and -0.95116 values between -0.95 and -0.80198 values between -0.80 and -0.64319 values between -0.64 and -0.49439 values between -0.49 and -0.33645 values between -0.33 and -0.18909 values between -0.18 and -0.02778 values between -0.02 and 0.13199 values between 0.13 and 0.2952 values between 0.29 and 0.4410 values between 0.44 and 0.601 values between 0.60 and 0.751 values between 0.75 and 0.913 values between 0.91 and 1.062 values between 1.22 and 1.372 values between 1.37 and 1.531 values between 1.68 and 1.841 values between 1.84 and 1.991 values between 2.46 and 2.61offnegativepositiveoften

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. Terminator
  2. 耳
  3. Expected
  4. ериал
  5. Format
  6. Paren
  7. ensued
  8. ilities
  9. implies
  10. Album

Suppresses

  1. centrally
  2. 衷
  3. ponsive
  4. 您好
  5. 踵
  6. DEX
  7. tures
  8. massage
  9. IMENT
  10. ⌠