Neuron Explorer

Qwen3-0.6B

separator in keyword and metadata lists

A neuron in layer 13 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 responds to small separators like commas, bars, bullets and slashes in lists of keywords, bylines and dates at the top of a document. The pattern is loose and mixed. Turning it up produced replies that were off topic, so the effect is unclear.

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

    Cell Biology⏎Chlamydomonas; Dynein ATPase; Flagella; Microscopy, Electron; Microtubules⏎Analysis of serial cross-sections of the Ch

  2. #2 peak

    Lets<|endoftext|>Jonathan D. Jaipaul, Devya Hemraj & Samantha Providence-Forrester⏎Published: November 4, 2020 • Book of Abstracts – Student

  3. #3 peak

    number words as they follow directions on this cute worksheet. Other freebies are available to download at: www.creativeclassroomconnections.com.⏎Freebies , FREE, Pre-K and Kindergarten Fine Motor Skills.⏎Days of the Week Worksheets⏎Apple

  4. #4 peak

    interpreting the observed movements. Ideas for future research to studying tuna behaviour near FADs are discussed.⏎Key words: tuna / Fish Aggregating Devices / behaviour / horizontal movements / ultrasonic telemetry / French Polynesia⏎© Elsevier,

  5. #5 peak

    process validation, lack of detailed knowledge about the interaction between high pressure, and a number of food constituents, packaging and statutory issues.⏎Keywords high pressure, food processing, non-thermal processing⏎Consumers demand high quality and convenient products with natural flavor

  6. #6 peak

    ivut||678 - 687|⏎|Julkaisu||JOURNAL OF LANGUAGE TEACHING AND RESEARCH|⏎|DOI - pysyväislinkit|⏎|Tila||Julkaistu -

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 want to take a look at the problem, but I can't seem to figure out how to make sense of the concept of the problem. I'm not sure. I realize that I've come across this (the problem) — I

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

    abola is the set of all points in a plane that are equidistant from the focus and the directrix of the parabola. [Adopted from iCoachMath]⏎Focus:Focus of a parabola lies on the

  2. #2 peak

    ampling; Paint-spraying; Sampling-methods; Sampling-equipment; Filtration; Analytical-methods; Equipment-design⏎Microbiology Colorado State University Department of Microbiology Fort Collins, Colo 80523⏎American Industrial

  3. #3 peak

    of a collection of Late Byzantine Mosaic Icons… but there is no proof for this. https://www.academia.edu/39903340/Displaying_an_Icon_The_Mosaic_Icon_of_S

  4. #4 peak

    ulkaisu||JOURNAL OF LANGUAGE TEACHING AND RESEARCH|⏎|DOI - pysyväislinkit|⏎|Tila||Julkaistu - 1 syysk. 2021|⏎|OKM

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. L10:36 excites
  2. L12:3037 excites
  3. L10:351 weakly suppresses
  4. L11:851 weakly suppresses
  5. L5:2313 weakly excites
  6. L10:1274 weakly excites
  7. L9:1496 weakly excites
  8. L5:629 weakly excites
  9. L8:311 weakly excites
  10. L12:1166 weakly suppresses
  11. L12:150 weakly suppresses
  12. L6:3049 weakly excites
  13. L12:73 weakly excites
  14. L1:1733 weakly excites
  15. L10:364 weakly suppresses
  16. L10:2376 weakly suppresses
  17. L8:554 weakly excites
  18. L10:471 weakly excites
  19. L10:1124 weakly suppresses
  20. L12:3029 weakly excites
  21. L2:737 weakly excites
  22. L9:2357 weakly excites
  23. L6:234 weakly suppresses
  24. L12:2948 weakly suppresses

Feeds later layers

  1. L19:954 weakly excites
  2. L17:77 weakly suppresses
  3. L15:603 weakly suppresses
  4. L18:370 weakly excites
  5. L17:293 weakly suppresses
  6. L16:1968 weakly excites
  7. L16:509 weakly excites
  8. L15:1794 weakly suppresses
  9. L18:904 weakly suppresses
  10. L16:439 weakly suppresses
  11. L27:160 weakly excites
  12. L18:263 weakly excites
  13. L20:494 weakly suppresses
  14. L18:1451 weakly excites
  15. L16:775 weakly suppresses
  16. L16:2677 weakly excites
  17. L17:1657 weakly excites
  18. L16:297 weakly excites
  19. L17:2179 weakly suppresses
  20. L16:2203 weakly excites
  21. L17:112 weakly suppresses
  22. L14:741 weakly suppresses
  23. L18:1673 weakly suppresses
  24. L17:2033 weakly excites

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.

3 values between -1.10 and -0.982 values between -0.98 and -0.861 values between -0.86 and -0.746 values between -0.74 and -0.625 values between -0.62 and -0.5027 values between -0.50 and -0.3894 values between -0.38 and -0.26317 values between -0.26 and -0.141030 values between -0.14 and -0.021526 values between -0.02 and 0.10618 values between 0.10 and 0.22183 values between 0.22 and 0.3450 values between 0.34 and 0.4715 values between 0.47 and 0.594 values between 0.59 and 0.716 values between 0.71 and 0.833 values between 0.83 and 0.952 values between 0.95 and 1.072 values between 1.07 and 1.191 values between 1.19 and 1.313 values between 1.43 and 1.551 values between 1.67 and 1.792 values between 1.91 and 2.032 values between 2.03 and 2.152 values between 3.12 and 3.242 values between 3.48 and 3.601 values between 3.60 and 3.72offnegativepositiveoften

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. offs
  2. bose
  3. theon
  4. olia
  5. GetType
  6. disposable
  7. iating
  8. luxe
  9. 远景
  10. @",

Suppresses

  1. because
  2. 因为我们
  3. 或者其他
  4. 因为在
  5. ල
  6. RCT
  7. because
  8. _PROCESS
  9. candidates
  10. 愿