Neuron Explorer

Qwen3-0.6B

full stops ending informative sentences

A neuron in layer 17 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 lights up on the full stop at the end of a factual sentence in explanatory writing, such as an article or guide. Turning it up made the model's output fall into loops of stock phrases like "I have I have" or "The Following", so the steering result tells us little about the neuron's meaning.

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

    sense, to conflate image production with graphic design. For example, contributors to WikiPedia often edit and re-upload images used to illustrate articles on that site. But in another way, it may not. Cropping or resizing a photograph is

  2. #2 peak

    cryptocurrencies?⏎For starters, cryptocurrencies are essentially digital assets that exist only in the digital realm. They do not exist in a physical form, i.e. paper or coins,

  3. #3 peak

    login required)<|endoftext|>California Protected Classes: What Are They?⏎In the landscape of employment and anti-discrimination laws, the concept of protected classes is paramount in California. These

  4. #4 peak

    in the following ways:⏎The cloud uses encryption techniques to protect your business data. Encryption converts your information into a secret code that can only be understood by authorised individuals. This helps prevent unauthorised access and keeps your data secure.⏎Cloud service providers

  5. #5 peak

    practice of⏎liberalism.ABOUT THE SERIES: The Very Short Introductions series from Oxford University Press contains hundreds of titles in almost every subject area. These pocket-sized books are the perfect way to get ahead in a new subject quickly

  6. #6 peak

    to adjust to the rigorous academic demands of college.⏎4. Obtaining Marketable Skills⏎Most employers today put more emphasis on graduates' skills than on their academic majors. Encourage your son or daughter to develop strengths in at least two or three of

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: and that I have been writing for a year. on a year of work for which I have been in charge of writing for... in the work and which I have been doing at work for... and that I have done on work at work

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

    workplace, HEPA filters, UV lights, laminar flow cabinets, and other biological safety cabinet components protect.⏎Read More @ https://cmibloginsight.blogspot.com/2023/05/biological-safety-cabinets

  2. #2 peak

    for use in K-12 lesson plans, visit the IHRC’s portal for digital resources via the University of Minnesota’s UMedia Archive:http://ihrc.umn.edu/research/digitalsources.php⏎Click here for additional resources.⏎

  3. #3 peak

    1863). International Journal on Stereo & Immersive Media, 2(2), 54-69. Retrieved from https://revistas.ulusofona.pt/index.php/stereo/article/view/6807<|endoftext|>

  4. #4 peak

    wrote my master's thesis on an extension to the hash join. You can view a copy of this thesis on my old school's Web site (http://www.cs.ndsu.nodak.edu/~peasland/paper.doc). If you

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. L16:432 strongly suppresses period after month or initial
  2. L11:132 excites period ending an opening topic sentence
  3. L15:403 suppresses period inside an abbreviation
  4. L14:422 excites periods after initials and abbreviations
  5. L4:7 excites
  6. L15:313 excites decimal point inside numbers
  7. L13:53 excites period ending a claim before elaboration
  8. L5:168 suppresses
  9. L5:155 suppresses
  10. L8:541 excites
  11. L8:143 suppresses
  12. L9:16 suppresses unclear: periods inside numbers and names
  13. L14:155 excites sentence-ending periods in prose
  14. L16:240 suppresses comma after 'sometimes' or 'the truth'
  15. L6:305 suppresses
  16. L9:2447 excites
  17. L16:1545 weakly excites
  18. L16:359 weakly excites
  19. L5:336 weakly suppresses
  20. L9:613 weakly suppresses
  21. L0:1670 weakly excites
  22. L2:9 weakly suppresses sentence-final periods vs dots in code
  23. L15:804 weakly suppresses
  24. L0:1703 weakly suppresses

Feeds later layers

  1. L21:1086 excites end of sentence before a claim or statistic
  2. L18:791 excites
  3. L19:384 excites
  4. L19:344 excites periods ending explanations
  5. L24:353 excites decimal point inside a number
  6. L20:962 excites decimal points and dots in numbers
  7. L22:67 excites
  8. L25:198 excites
  9. L27:133 excites
  10. L20:505 excites
  11. L21:524 excites
  12. L23:179 excites
  13. L23:509 excites
  14. L25:217 excites
  15. L19:68 excites full stops closing a claim
  16. L27:640 excites 'of' inside fixed titles and names
  17. L19:1072 excites
  18. L19:2097 excites sentence-end punctuation in speech
  19. L20:473 excites
  20. L23:177 excites
  21. L20:940 excites
  22. L19:330 excites
  23. L20:296 excites
  24. L18:126 suppresses comma joining a full clause

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 -7.83 and -6.751 values between -5.68 and -4.601 values between -4.60 and -3.532 values between -3.53 and -2.4514 values between -2.45 and -1.382063 values between -1.38 and -0.301703 values between -0.30 and 0.7827 values between 0.78 and 1.856 values between 1.85 and 2.934 values between 2.93 and 4.005 values between 4.00 and 5.083 values between 5.08 and 6.152 values between 6.15 and 7.231 values between 8.30 and 9.381 values between 10.45 and 11.532 values between 11.53 and 12.602 values between 12.60 and 13.681 values between 13.68 and 14.762 values between 14.76 and 15.831 values between 15.83 and 16.913 values between 16.91 and 17.985 values between 17.98 and 19.062 values between 19.06 and 20.137 values between 20.13 and 21.2111 values between 21.21 and 22.286 values between 22.28 and 23.366 values between 23.36 and 24.435 values between 24.43 and 25.5112 values between 25.51 and 26.582 values between 26.58 and 27.663 values between 27.66 and 28.741 values between 29.81 and 30.891 values between 30.89 and 31.961 values between 31.96 and 33.041 values between 34.11 and 35.19offnegativepositiveoften

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. However
  2. о
  3. Currently
  4. Consider
  5. ⏎⏎
  6. бег
  7. Present
  8. 地上
  9. Simple
  10. computations

Suppresses

  1. .Template
  2. -disable
  3. 化的
  4. 全文
  5. portion
  6. .]
  7. zelf
  8. igkeit
  9. .Bar
  10. .Dep