Neuron Explorer

Qwen3-0.6B

immediately repeated words

A neuron in layer 8 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron tends to fire on the second copy of a word that has just been repeated, like 'and and' or 'could could'. Several of the strongest examples are this kind of stutter. Turning it up did not bring out a clear theme in the test 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

    triangles to reduce the amount of steel needed for construction by over 20%.⏎New computer technology is allowing architects to explore radically redesigned structures, with curves and and angles rather than the traditional rectilinear form. The same technology enables the designer to

  2. #2 peak

    with adequate wall thickness, (2) how to integrated the speaker grille into the plaster form, and (3) creating a suitable mold for the plaster that could could be re-used and not damage the part when removed.⏎With regards to the first

  3. #3 peak

    of relative mobility between states, but there are some important factors that the Economic Mobility Project has found are essential to promoting economic mobility overall, and these include things like like “educational attainment, savings and asset building, and neighborhood poverty during childhood

  4. #4 peak

    for scores of Australians. This has had dentists all around the country looking for ways to increase awareness about sound oral hygiene habits as a means to limit the number or extractions, cavities and instances of gum disease in young as well as senior

  5. #5 peak

    , animals and/or insects may use them for habitat or food. When natural areas are lost or cut into pieces or harmed in other ways from our activities, what' s left becomes all the more critical for the survival of many species. By using

  6. #6 peak

    ⏎- Tumors of chest wall⏎- In newborns, high levels of circulating estrogen may result in enlarged breast tissue and secretion of milk⏎Galactorrhea is more common in women. Other factors that may increase your chance of gal

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: "What is the relationship between the number of years, the number of parameters, and the correlation coefficient?" Let我开始思考这些数据如何影响结果。 可能的参数是哪些? 现在开始分析这些数据如何影响结果。 首先,

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

    be asked to critically think and debate varying interpretations of the empirical evidence in relation to the science being brought to your attention<|endoftext|>Economics– Important Questions Bank for Rajasthan Board HSC 201

  2. #2 peak

    what you all are waiting for. Teachers, parents and student, now you will find below the Solutions of Class 10 Science Chapter 4 Environmental management questions answers. From the subject English of Maharashtra State Board.⏎Below you can see

  3. #3 peak

    in a step by step manner. All these solutions are beneficial for the students of class 10 appearing for the board exams. Students can easily NCERT Solutions for Class 10 Maths Exercise 6.5 Chapter 6 Triangles from

  4. #4 peak

    6 Examination.⏎We had mentioned some tips for cracking the HSC Rajasthan Board exam here: HSC Study Tips to Crack HSC Exams.⏎We had

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. L3:1016 excites
  2. L7:1216 excites
  3. L6:2237 excites
  4. L6:1021 suppresses
  5. L7:2270 suppresses "of" after member, version, or part nouns
  6. L5:2305 suppresses
  7. L4:561 weakly excites
  8. L7:2243 weakly excites
  9. L4:385 weakly excites multiplication and scaling in math
  10. L2:1389 weakly suppresses
  11. L1:1029 weakly excites
  12. L3:1138 weakly excites
  13. L3:1236 weakly excites
  14. L2:1437 weakly suppresses
  15. L0:83 weakly suppresses
  16. L2:1026 weakly suppresses
  17. L7:2919 weakly excites
  18. L0:1444 weakly excites
  19. L4:405 weakly suppresses
  20. L3:529 weakly suppresses
  21. L4:2030 weakly excites
  22. L3:827 weakly suppresses
  23. L1:1418 weakly suppresses
  24. L5:409 weakly suppresses

Feeds later layers

  1. L9:2817 excites
  2. L10:718 excites
  3. L15:1198 excites
  4. L11:1393 excites
  5. L17:1968 excites
  6. L10:2625 excites
  7. L9:2354 weakly excites
  8. L18:2205 weakly excites
  9. L9:752 weakly excites
  10. L19:1887 weakly excites
  11. L23:775 weakly suppresses
  12. L27:1195 weakly suppresses
  13. L10:77 weakly excites
  14. L16:1163 weakly suppresses
  15. L19:184 weakly suppresses
  16. L18:580 weakly excites
  17. L10:1263 weakly suppresses
  18. L14:534 weakly suppresses
  19. L13:1045 weakly suppresses
  20. L19:2615 weakly excites
  21. L10:2282 weakly excites
  22. L24:2936 weakly excites
  23. L9:1456 weakly suppresses
  24. L21:1637 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.

4 values between -0.49 and -0.437 values between -0.43 and -0.3624 values between -0.36 and -0.3066 values between -0.30 and -0.23142 values between -0.23 and -0.17330 values between -0.17 and -0.10538 values between -0.10 and -0.04812 values between -0.04 and 0.03732 values between 0.03 and 0.09570 values between 0.09 and 0.16374 values between 0.16 and 0.22179 values between 0.22 and 0.2972 values between 0.29 and 0.3535 values between 0.35 and 0.4217 values between 0.42 and 0.483 values between 0.48 and 0.551 values between 0.61 and 0.681 values between 1.40 and 1.461 values between 2.05 and 2.11offnegativepositiveoften

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. リア
  2. ault
  3. mankind
  4. {.
  5. フリー
  6. plaintext
  7. 러
  8. venture
  9. dere
  10. cuales

Suppresses

  1. 柬
  2. 区管委会
  3. éra
  4. 属
  5. VK
  6. ensburg
  7. CHR
  8. BOARD
  9. eworld
  10. ycz