Neuron Explorer

Qwen3-0.6B

'that', 'about' or 'of' after a verb or noun

A neuron in layer 3 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on small connecting words such as 'that', 'about' or 'of' right after a verb or noun, where a clause or topic is about to start, like 'warned of' or 'heard about'. Turning it up did not produce a clear, related change in the 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

    . All rights reserved. URL:http://www.healthscout.com/template.asp?id=669525<|endoftext|>It is certain, in any case, that ignorance, allied with power, is the most ferocious enemy justice can have.⏎

  2. #2 peak

    see there are so many questions asked meanwhile and students are expected to answers them. Some students may face some problem while answering it , but we promise you that after reviewing our answer, you will gain the confidence to answers such questions.

  3. #3 peak

    about television. Radio manufacturers rightfully worried that if people started buying TVs en masse, why would anyone want a new radio?⏎As Albert Abramson explains in his book about the early history of TV, demonstrations of TV technology were explicitly banned from the

  4. #4 peak

    can only do some rough estimation."⏎The Czech Republic - like Central Europe as a whole - is in a far better position than countries further east. But officials warn of a dangerous paradox: because the number of registered cases is so low, people become

  5. #5 peak

    on CNC machines, we had to keep the geometry simple because complex shapes would greatly increase manufacturing costs," says WHOI Supervising Engineer Bob McCabe." Then we heard about FDM and its ability to produce polycarbonate or ABS parts with the strength

  6. #6 peak

    was a blow to United States pride and prestige," reported the Christian Science Monitor. Time magazine quipped that Project Vanguard should be renamed Project Rearguard, and jokes about the botched

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 have a new. I have a new. I have a new. I have a new. I have a new. I have a new. I have a new. I have a new. I have a new. I have a

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

    08, OR<|endoftext|>A Global Index of Sustainable Well-Being⏎Source: The New Economics Foundation⏎From the Executive Summary:⏎There is a growing global consensus that we need new measures of progress. It is critical that these measures clearly reflect what

  2. #2 peak

    of his articles.⏎This narrow view drives some of those who cast a suspicious glance at the writer living outside of his country to end up with the naïve idea that it must be hard for writers in exile to write about their home countries: it

  3. #3 peak

    legends point out to various periods between this interval. Of course, some people might have arrived later and others earlier, but it is generally accepted by the great public that the island was uninhabited before 300 BC,

  4. #4 peak

    is missing or abducted and there is reason to⏎believe the child is in imminent danger, and there is information⏎available to disseminate to the general public that could assist in the⏎safe recovery of the child and/or the apprehension of

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. L2:651 suppresses
  2. L2:1336 excites
  3. L2:995 weakly excites
  4. L2:2505 weakly excites
  5. L2:406 weakly excites
  6. L1:1095 weakly suppresses
  7. L1:2568 weakly excites
  8. L2:683 weakly excites
  9. L1:436 weakly suppresses
  10. L2:41 weakly suppresses
  11. L1:1520 weakly excites
  12. L2:954 weakly excites
  13. L1:1137 weakly suppresses
  14. L1:227 weakly excites
  15. L0:1003 weakly suppresses
  16. L1:2306 weakly excites
  17. L2:618 weakly excites
  18. L2:1824 weakly excites
  19. L1:749 weakly excites
  20. L2:1095 weakly excites
  21. L1:2399 weakly suppresses
  22. L1:470 weakly excites
  23. L2:2043 weakly suppresses
  24. L0:2515 weakly suppresses

Feeds later layers

  1. L5:436 suppresses
  2. L4:223 suppresses
  3. L5:1949 suppresses
  4. L4:1349 weakly suppresses
  5. L19:1884 weakly excites
  6. L5:1043 weakly suppresses
  7. L14:1402 weakly excites
  8. L17:232 weakly suppresses
  9. L5:522 weakly excites
  10. L6:2450 weakly suppresses 'guess' and 'obvious' after 'as you can see'
  11. L5:165 weakly excites
  12. L5:724 weakly suppresses
  13. L5:1741 weakly excites
  14. L9:597 weakly suppresses
  15. L5:473 weakly suppresses
  16. L5:438 weakly suppresses
  17. L5:1297 weakly suppresses
  18. L7:1057 weakly suppresses
  19. L14:262 weakly suppresses
  20. L10:923 weakly excites
  21. L6:229 weakly suppresses
  22. L20:1510 weakly suppresses
  23. L12:2653 weakly excites
  24. L5:2133 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 -4.05 and -3.911 values between -2.94 and -2.801 values between -2.80 and -2.662 values between -2.52 and -2.381 values between -2.24 and -2.104 values between -1.82 and -1.682 values between -1.68 and -1.542 values between -1.54 and -1.403 values between -1.40 and -1.263 values between -1.26 and -1.121 values between -1.12 and -0.9810 values between -0.98 and -0.859 values between -0.85 and -0.719 values between -0.71 and -0.5710 values between -0.57 and -0.4353 values between -0.43 and -0.29253 values between -0.29 and -0.151261 values between -0.15 and -0.011572 values between -0.01 and 0.13436 values between 0.13 and 0.27149 values between 0.27 and 0.4178 values between 0.41 and 0.5528 values between 0.55 and 0.6911 values between 0.69 and 0.833 values between 0.83 and 0.973 values between 0.97 and 1.111 values between 1.25 and 1.391 values between 1.39 and 1.52offnegativepositiveoften

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. erville
  2. trand
  3. pires
  4. .onStart
  5. ler
  6. uin
  7. aping
  8. thood
  9. .getService
  10. 嫌疑

Suppresses

  1. ikal
  2. ucken
  3. DEAL
  4. град
  5. 谩
  6. deal
  7. Aston
  8. MADE
  9. shake
  10. pictures