Neuron Explorer

Qwen3-0.6B

'conf' word fragment (confine, confederation)

A neuron in layer 26 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 fires on the piece 'conf' at the start of words like confine, confederation and conferring. It also fires on a few name fragments like 'Duf' in a French name. Turning it up gave wandering, loosely connected text.

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

    issued from Barrelon workshop. In fact, they are still there.⏎Georges Dufêtre who has been one of Barrelon employees (since lst February 18

  2. #2 peak

    , researchers from HSPH, Bristol-Myers Squibb, and elsewhere describe how a designer compound mimics many of these protective effects in mice, conferring substantial immunity to diabetes, heart disease, and other metabolic problems. This immunity

  3. #3 peak

    -sur-Dargoire (1856)…⏎By 19th June 1878; when one of his employees named Georges Nicolas Dufêtre took over the workshop, Barrelon house extended the sphere of activity. In

  4. #4 peak

    the generation of on-demand photon pairs that are entangled. The research also integrates the materials with chip-scale silicon photonic devices that confine the light to extremely small volume to enhance the light emission efficiency. Ultimately,

  5. #5 peak

    Confederacy:⏎Article II. Each state retains its sovereignty, freedom, and independence, and every Power, Jurisdiction, and right, which is not by this confederation expressly delegated to the United States, in Congress assembled.⏎Provide for common defence

  6. #6 peak

    also taken into consideration.<|endoftext|>Family Disaster Plan⏎Make A Plan⏎Disaster can strike quickly and without warning. It can force you to evacuate your neighborhood or confine you to your home. What would you do if basic services were cut off

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: records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records records

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

    occasion of the State Opening of Parliament when travelling to and fro.⏎What Is Significant About the Designs of the Coin?⏎The effigy of Her Majesty wearing the Diamond Diadem featured on Canadian circulation coins from 1

  2. #2 peak

    , the obverse features a previous effigy of Her Majesty wearing the magnificent Diamond Diadem, while the reverse features a maple leaf motif as a symbol of Canada.⏎

  3. #3 peak

    The ammonium concentrations found in a buddingtonite sample had an RSD of 2.2% and their mean agreed with that obtained by the Kjeldahl method. By use of the prescribed dilution of the sample, detection limits

  4. #4 peak

    repertory.⏎The Symphony might be regarded as a battle, a study of opposites in conflict: the stately versus the agitated, the singable diatonic melody against the sinewy chromatic theme, the foreboding

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. L3:133 suppresses
  2. L5:2 suppresses
  3. L7:3 suppresses
  4. L7:287 suppresses
  5. L8:2395 weakly excites
  6. L13:620 weakly suppresses
  7. L9:18 weakly suppresses text boundary after a document break
  8. L22:115 weakly suppresses
  9. L1:2877 weakly excites
  10. L19:359 weakly suppresses
  11. L22:369 weakly excites
  12. L10:32 weakly excites
  13. L0:1369 weakly suppresses
  14. L23:389 weakly suppresses
  15. L20:429 weakly excites
  16. L5:135 weakly suppresses
  17. L18:541 weakly suppresses
  18. L2:1436 weakly excites
  19. L2:1611 weakly excites
  20. L6:864 weakly excites
  21. L9:26 weakly excites
  22. L17:1581 weakly excites
  23. L19:299 weakly excites
  24. L20:133 weakly suppresses

Feeds later layers

  1. L27:2270 weakly suppresses
  2. L27:3071 weakly suppresses
  3. L27:2916 weakly excites
  4. L27:2059 weakly suppresses
  5. L27:1385 weakly excites
  6. L27:1090 weakly suppresses
  7. L27:2218 weakly excites
  8. L27:1950 weakly suppresses
  9. L27:1030 weakly excites
  10. L27:330 weakly excites
  11. L27:1213 weakly suppresses
  12. L27:1071 weakly excites
  13. L27:496 weakly excites
  14. L27:2424 weakly suppresses
  15. L27:728 weakly excites
  16. L27:2656 weakly suppresses
  17. L27:2918 weakly suppresses
  18. L27:2836 weakly excites
  19. L27:837 weakly excites
  20. L27:2500 weakly excites
  21. L27:2653 weakly excites
  22. L27:2894 weakly excites
  23. L27:1833 weakly excites
  24. L27:2090 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.

1 values between -26.44 and -25.471 values between -14.84 and -13.871 values between -13.87 and -12.902 values between -12.90 and -11.941 values between -11.94 and -10.973 values between -10.00 and -9.041 values between -9.04 and -8.077 values between -8.07 and -7.107 values between -7.10 and -6.145 values between -6.14 and -5.172 values between -5.17 and -4.2011 values between -4.20 and -3.2317 values between -3.23 and -2.2724 values between -2.27 and -1.30201 values between -1.30 and -0.333254 values between -0.33 and 0.63345 values between 0.63 and 1.608 values between 1.60 and 2.576 values between 2.57 and 3.532 values between 3.53 and 4.503 values between 4.50 and 5.473 values between 6.43 and 7.401 values between 7.40 and 8.371 values between 8.37 and 9.331 values between 11.27 and 12.23offnegativepositiveoften

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. heim
  2. żenia
  3. roit
  4. ucid
  5. عال
  6. olar
  7. ides
  8. adro
  9. зд
  10. Lİ

Suppresses

  1. 国务
  2. resources
  3. 笑脸
  4. 自然资源
  5. 昴
  6. resource
  7. ALLY
  8. 誕
  9. shack
  10. Airbus