Neuron Explorer

Qwen3-0.6B

slot before the second number in a range

A neuron in layer 20 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 the spot right before the second number in a pair or range, such as the gap between '80' and '20' in '80% and 20%' or '80-90%'. Its negative side fires inside numbers, on a middle digit. Turning it up made the replies wander without a clear theme.

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

    wood, plaster, clay, or other substance to give it the colour and lustre of bronze. Dutch metal, an alloy of 80 percent copper and 20 percent zinc, is frequently used for bronzing.

  2. #2 peak

    -lash type injury. No imaging can identify a concussion and it is purely assessed on signs and symptoms that occur immediately or within a few days following.⏎80-90% of people with a concussion make a complete recovery within 1-

  3. #3 peak

    40% lean protein, 20% fat and 40% complex carbohydrates for optimum nutrition and satisfaction. One of the best things

  4. #4 peak

    ant/mood-stabilizing medication than the one you previously selected (other than lithium) that a psychiatrist might prescribe to treat bipolar disorders.⏎300 to 50

  5. #5 peak

    for the entire class.⏎PC: Windows XP and up⏎Mac: OS X (10.1) and up⏎Resolution: 800 x 600 or higher<|endoftext|>In response to the theme of 201

  6. #6 peak

    local wind shear was having its effect on the storm, and Hurricane Humberto’s winds had begun to diminish. By 1500 UTC (11:00 a.m), Humberto had weakened to tropical storm status, with maximum

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: the 0000000000000000000000000000000000000000000000

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

    the early 1900s. The construction of the hotel began in 1901 and was completed in just two years, in 1903. The hotel was designed by Harrison Albright, an architect known for

  2. #2 peak

    mm) – this is one thousandth of a meter. In scientific notation, 1 millimeter is 1 – E3 m, which means 0.001 x 1

  3. #3 peak

    Consortium (CFSC) developed a three year C4E initiative (2007-2010) to pilot test the approach and tools articulated in the Practical Guidance Note

  4. #4 peak

    At the end of 1099 he visited Jerusalem along with Bohemund I of Antioch, but he returned to Edessa in January 1100. After Godfrey's death in 1100 he

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. L17:1423 excites
  2. L19:2772 suppresses
  3. L16:2181 excites
  4. L17:1622 suppresses
  5. L17:1552 suppresses
  6. L18:1781 weakly excites
  7. L17:2082 weakly suppresses
  8. L19:1284 weakly suppresses digits inside years and decades
  9. L18:2858 weakly suppresses
  10. L19:616 weakly excites
  11. L13:2726 weakly excites
  12. L3:2926 weakly excites
  13. L19:2663 weakly excites
  14. L1:2404 weakly suppresses
  15. L17:107 weakly suppresses
  16. L13:2047 weakly suppresses
  17. L7:22 weakly suppresses
  18. L9:346 weakly suppresses
  19. L19:1851 weakly excites
  20. L9:42 weakly suppresses
  21. L8:2791 weakly suppresses
  22. L19:619 weakly suppresses
  23. L17:692 weakly suppresses
  24. L1:458 weakly suppresses

Feeds later layers

  1. L22:402 excites
  2. L23:775 suppresses
  3. L21:1732 suppresses
  4. L25:621 suppresses
  5. L23:2924 weakly suppresses
  6. L21:1053 weakly excites
  7. L25:1020 weakly suppresses
  8. L21:2918 weakly suppresses
  9. L24:797 weakly suppresses
  10. L22:1203 weakly suppresses
  11. L21:2428 weakly excites
  12. L21:972 weakly suppresses
  13. L25:492 weakly suppresses
  14. L22:2783 weakly suppresses
  15. L21:452 weakly excites
  16. L21:1182 weakly suppresses
  17. L23:1128 weakly suppresses
  18. L21:719 weakly excites
  19. L25:1662 weakly suppresses
  20. L23:1945 weakly suppresses
  21. L23:1411 weakly suppresses
  22. L26:751 weakly excites
  23. L24:992 weakly suppresses
  24. L22:1319 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 -12.28 and -11.581 values between -10.17 and -9.471 values between -5.96 and -5.261 values between -5.26 and -4.561 values between -4.56 and -3.851 values between -3.85 and -3.154 values between -3.15 and -2.454 values between -2.45 and -1.7511 values between -1.75 and -1.0446 values between -1.04 and -0.343400 values between -0.34 and 0.36391 values between 0.36 and 1.0613 values between 1.06 and 1.779 values between 1.77 and 2.472 values between 2.47 and 3.172 values between 3.17 and 3.873 values between 3.87 and 4.582 values between 4.58 and 5.283 values between 5.28 and 5.982 values between 5.98 and 6.681 values between 6.68 and 7.381 values between 7.38 and 8.091 values between 8.09 and 8.793 values between 8.79 and 9.491 values between 9.49 and 10.191 values between 10.19 and 10.901 values between 12.30 and 13.001 values between 15.11 and 15.81offnegativepositiveoften

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. igo
  2. LOSS
  3. >Type
  4. 蒟
  5. truth
  6. predic
  7. gettext
  8. FormatException
  9. (det
  10. 阵地

Suppresses

  1. rt
  2. iros
  3. v
  4. _v
  5. sz
  6. ствовать
  7. JD
  8. ulers
  9. 焉
  10. irable